Subversion Repositories tendra.SVN

Rev

Rev 2 | Details | Compare with Previous | Last modification | View Log | RSS feed

Rev Author Line No. Line
2 7u83 1 
<!-- Crown Copyright (c) 1998 -->
2 
<HTML>
3 
<HEAD>
4 
<TITLE>
5 
C++ Producer Guide: Implementation
6 
</TITLE>
7 
</HEAD>
8 
<BODY TEXT="#000000" BGCOLOR="#FFFFFF" LINK="#0000FF" VLINK="#400080" ALINK="#FF0000">
9 
 
10 
<H1>C++ Producer Guide</H1>
11 
<H3>March 1998</H3>
12 
<A HREF="std.html"><IMG SRC="../images/next.gif" ALT="next section"></A>
13 
<A HREF="link.html"><IMG SRC="../images/prev.gif" ALT="previous section"></A>
14 
<A HREF="index.html"><IMG SRC="../images/top.gif" ALT="current document"></A>
15 
<A HREF="../index.html"><IMG SRC="../images/home.gif" ALT="TenDRA home page">
16 
</A>
17 
<IMG SRC="../images/no_index.gif" ALT="document index"><P>
18 
<HR>
19 
 
20 
<DL>
21 
<DT><A HREF="#arith"><B>2.6.1</B> - Arithmetic types</A><DD>
22 
<DT><A HREF="#literal"><B>2.6.2</B> - Integer literal types</A><DD>
23 
<DT><A HREF="#bitfield"><B>2.6.3</B> - Bitfield types</A><DD>
24 
<DT><A HREF="#pointer"><B>2.6.4</B> - Generic pointers</A><DD>
25 
<DT><A HREF="#conv"><B>2.6.5</B> - Undefined conversions</A><DD>
26 
<DT><A HREF="#div"><B>2.6.6</B> - Integer division</A><DD>
27 
<DT><A HREF="#call"><B>2.6.7</B> - Calling conventions</A><DD>
28 
<DT><A HREF="#ptr_mem"><B>2.6.8</B> - Pointers to data members</A><DD>
29 
<DT><A HREF="#ptr_mem_func"><B>2.6.9</B> - Pointers to function members</A><DD>
30 
<DT><A HREF="#class"><B>2.6.10</B> - Class layout</A><DD>
31 
<DT><A HREF="#derive"><B>2.6.11</B> - Derived class layout</A><DD>
32 
<DT><A HREF="#constr"><B>2.6.12</B> - Constructors and destructors</A><DD>
33 
<DT><A HREF="#vtable"><B>2.6.13</B> - Virtual function tables</A><DD>
34 
<DT><A HREF="#rtti"><B>2.6.14</B> - Run-time type information</A><DD>
35 
<DT><A HREF="#init"><B>2.6.15</B> - Dynamic initialisation</A><DD>
36 
<DT><A HREF="#except"><B>2.6.16</B> - Exception handling</A><DD>
37 
<DT><A HREF="#mangle"><B>2.6.17</B> - Mangled identifier names</A><DD>
38 
</DL>
39 
<HR>
40 
 
41 
<H2>2.6. Implementation details</H2>
42 
<P>
43 
This section describes various of the implementation details of the
44 
C++ producer TDF output.  In particular it describes the standard
45 
TDF tokens used to represent the target dependent aspects of the language
46 
and to provide links into the run-time system.  Many of these tokens
47 
are common to the C and C++ producers.  Those which are unique to
48 
the C++ producer have names of the form <CODE>~cpp.*</CODE>.  Note
49 
that the description is in terms of TDF tokens, not the internal tokens
50 
introduced by the
51 
<A HREF="token.html"><CODE>#pragma token</CODE> syntax</A>.
52 
</P>
53 
<P>
54 
There are two levels of implementation in the run-time system.  The
55 
actual interface between the producer and the run-time system is given
56 
by the standard tokens.  The provided implementation defines these
57 
tokens in a way appropriate to itself.  An alternative implementation
58 
would have to define the tokens differently.  It is intended that
59 
the standard tokens are sufficiently generic to allow a variety of
60 
implementations to hook into the producer output in the manner they
61 
require.
62 
</P>
63 
 
64 
<HR>
65 
<H3><A NAME="arith">2.6.1. Arithmetic types</A></H3>
66 
<P>
67 
The representations of the basic arithmetic types are target dependent,
68 
so, for example, an <CODE>int</CODE> may contain 16, 32, 64 or some
69 
other number of bits.  Thus it is necessary to introduce a token to
70 
stand for each of the built-in arithmetic types (including the
71 
<A HREF="pragma.html#longlong"><CODE>long long</CODE> types</A>).
72 
Each integral type is represented by a <CODE>VARIETY</CODE> token
73 
as follows:
74 
</P>
75 
<CENTER>
76 
<TABLE BORDER>
77 
<TR><TH>Type</TH>
78 
<TH>Token</TH>
79 
<TH>Encoding</TH>
80 
<TR><TD ALIGN=CENTER>char</TD>
81 
<TD ALIGN=CENTER>~char</TD>
82 
<TD ALIGN=CENTER>0</TD>
83 
<TR><TD ALIGN=CENTER>signed char</TD>
84 
<TD ALIGN=CENTER>~signed_char</TD>
85 
<TD ALIGN=CENTER>0 | 4 = 4</TD>
86 
<TR><TD ALIGN=CENTER>unsigned char</TD>
87 
<TD ALIGN=CENTER>~unsigned_char</TD>
88 
<TD ALIGN=CENTER>0 | 8 = 8</TD>
89 
<TR><TD ALIGN=CENTER>signed short</TD>
90 
<TD ALIGN=CENTER>~signed_short</TD>
91 
<TD ALIGN=CENTER>1 | 4 = 5</TD>
92 
<TR><TD ALIGN=CENTER>unsigned short</TD>
93 
<TD ALIGN=CENTER>~unsigned_short</TD>
94 
<TD ALIGN=CENTER>1 | 8 = 9</TD>
95 
<TR><TD ALIGN=CENTER>signed int</TD>
96 
<TD ALIGN=CENTER>~signed_int</TD>
97 
<TD ALIGN=CENTER>2 | 4 = 6</TD>
98 
<TR><TD ALIGN=CENTER>unsigned int</TD>
99 
<TD ALIGN=CENTER>~unsigned_int</TD>
100 
<TD ALIGN=CENTER>2 | 8 = 10</TD>
101 
<TR><TD ALIGN=CENTER>signed long</TD>
102 
<TD ALIGN=CENTER>~signed_long</TD>
103 
<TD ALIGN=CENTER>3 | 4 = 7</TD>
104 
<TR><TD ALIGN=CENTER>unsigned long</TD>
105 
<TD ALIGN=CENTER>~unsigned_long</TD>
106 
<TD ALIGN=CENTER>3 | 8 = 11</TD>
107 
<TR><TD ALIGN=CENTER>signed long long</TD>
108 
<TD ALIGN=CENTER>~signed_longlong</TD>
109 
<TD ALIGN=CENTER>3 | 4 | 16 = 23 </TD>
110 
<TR><TD ALIGN=CENTER>unsigned long long</TD>
111 
<TD ALIGN=CENTER>~unsigned_longlong</TD>
112 
<TD ALIGN=CENTER>3 | 8 | 16 = 27</TD>
113 
</TABLE>
114 
</CENTER>
115 
<P>
116 
Similarly each floating point type is represent by a
117 
<CODE>FLOATING_VARIETY</CODE> token:
118 
</P>
119 
<CENTER>
120 
<TABLE BORDER>
121 
<TR><TH>Type</TH>   <TH>Token</TH>
122 
<TR><TD ALIGN=CENTER>float</TD>  <TD ALIGN=CENTER>~float</TD>
123 
<TR><TD ALIGN=CENTER>double</TD> <TD ALIGN=CENTER>~double</TD>
124 
<TR><TD ALIGN=CENTER>long double</TD> <TD ALIGN=CENTER>~long_double</TD>
125 
</TABLE>
126 
</CENTER>
127 
<P>
128 
Each integral type also has an encoding as a <CODE>SIGNED_NAT</CODE>
129 
as shown above.  This number is a bit pattern built up from the following
130 
values:
131 
</P>
132 
<CENTER>
133 
<TABLE BORDER>
134 
<TR><TH>Type</TH>   <TH>Encoding</TH>
135 
<TR><TD ALIGN=CENTER>char</TD>  <TD ALIGN=CENTER>0</TD>
136 
<TR><TD ALIGN=CENTER>short</TD>  <TD ALIGN=CENTER>1</TD>
137 
<TR><TD ALIGN=CENTER>int</TD>  <TD ALIGN=CENTER>2</TD>
138 
<TR><TD ALIGN=CENTER>long</TD>  <TD ALIGN=CENTER>3</TD>
139 
<TR><TD ALIGN=CENTER>signed</TD> <TD ALIGN=CENTER>4</TD>
140 
<TR><TD ALIGN=CENTER>unsigned</TD> <TD ALIGN=CENTER>8</TD>
141 
<TR><TD ALIGN=CENTER>long long</TD> <TD ALIGN=CENTER>16</TD>
142 
</TABLE>
143 
</CENTER>
144 
<P>
145 
Any target dependent integral type can be represented by a
146 
<CODE>SIGNED_NAT</CODE> token using this encoding.  This representation,
147 
rather than one based on <CODE>VARIETY</CODE>s, is used for ease of
148 
manipulation.  The token:
149 
<PRE>
150 
	~convert : ( SIGNED_NAT ) -&gt; VARIETY
151 
</PRE>
152 
gives the mapping from the integral encoding to the representing variety.
153 
For example, it will map <CODE>6</CODE> to <CODE>~signed_int</CODE>.
154 
</P>
155 
<P>
156 
The token:
157 
<PRE>
158 
	~promote : ( SIGNED_NAT ) -&gt; SIGNED_NAT
159 
</PRE>
160 
describes how to form the promotion of an integral type according
161 
to the ISO C/C++ value preserving rules, and is used by the producer
162 
to represent target dependent promotion types.  For example, the promotion
163 
of <CODE>unsigned short</CODE> may be <CODE>int</CODE> or <CODE>unsigned
164 
int</CODE> depending on the representation of these types; that is
165 
to say, <CODE>~promote ( 9 )</CODE> will be <CODE>6</CODE> on some
166 
machines and <CODE>10</CODE> on others.  Although <CODE>~promote</CODE>
167 
is used by default, a program may specify another token with the same
168 
sort signature to be used in its place by means of the directive:
169 
<PRE>
170 
	#pragma TenDRA compute promote <I>identifier</I>
171 
</PRE>
172 
For example, a standard token <CODE>~sign_promote</CODE> is defined
173 
which gives the older C sign preserving promotion rules.  In addition,
174 
the promotion of an individual type can be specified using:
175 
<PRE>
176 
	#pragma TenDRA promoted <I>type-id</I> : <I>promoted-type-id</I>
177 
</PRE>
178 
</P>
179 
<P>
180 
The token:
181 
<PRE>
182 
	~arith_type : ( SIGNED_NAT, SIGNED_NAT ) -&gt; SIGNED_NAT
183 
</PRE>
184 
similarly describes how to form the usual arithmetic result type from
185 
two promoted integral operand types.  For example, the arithmetic
186 
type of <CODE>long</CODE> and <CODE>unsigned int</CODE> may be
187 
<CODE>long</CODE> or <CODE>unsigned long</CODE> depending on the representation
188 
of these types; that is to say,
189 
<CODE>~arith_type ( 7, 10 )</CODE> will be <CODE>7</CODE> on some
190 
machines and <CODE>11</CODE> on others.
191 
</P>
192 
<P>
193 
Any tokenised type declared using:
194 
<PRE>
195 
	#pragma token VARIETY v # tv
196 
</PRE>
197 
will be represented by a <CODE>SIGNED_NAT</CODE> token with external
198 
name
199 
<CODE>tv</CODE> corresponding to the encoding of <CODE>v</CODE>.
200 
Special cases of this are the implementation dependent integral types
201 
which arise naturally within the language.  The external token names
202 
for these types are given below:
203 
</P>
204 
<CENTER>
205 
<TABLE BORDER>
206 
<TR><TH>Type</TH>   <TH>Token</TH>
207 
<TR><TD ALIGN=CENTER>bool</TD>  <TD ALIGN=CENTER>~cpp.bool</TD>
208 
<TR><TD ALIGN=CENTER>ptrdiff_t</TD> <TD ALIGN=CENTER>ptrdiff_t</TD>
209 
<TR><TD ALIGN=CENTER>size_t</TD> <TD ALIGN=CENTER>size_t</TD>
210 
<TR><TD ALIGN=CENTER>wchar_t</TD> <TD ALIGN=CENTER>wchar_t</TD>
211 
</TABLE>
212 
</CENTER>
213 
<P>
214 
So, for example, a <CODE>sizeof</CODE> expression has shape
215 
<CODE>~convert ( size_t )</CODE>.  The token <CODE>~cpp.bool</CODE>
216 
is defined in the default implementation, but the other tokens are
217 
defined according to their definitions on the target machine in the
218 
normal API library building mechanism.
219 
</P>
220 
 
221 
<HR>
222 
<H3><A NAME="literal">2.6.2. Integer literal types</A></H3>
223 
<P>
224 
The <A HREF="pragma.html#int">type of an integer literal</A> is defined
225 
in terms of the first in a list of possible integral types.  The first
226 
type in which the literal value can be represented gives the type
227 
of the literal.  For small literals it is possible to work out the
228 
type exactly, however for larger literals the result is target dependent.
229 
For example, the literal <CODE>50000</CODE> will have type <CODE>int</CODE>
230 
on machines in which <CODE>50000</CODE> fits into an <CODE>int</CODE>,
231 
and
232 
<CODE>long</CODE> otherwise.  This target dependent mapping is given
233 
by a series of tokens of the form:
234 
<PRE>
235 
	~lit_* : ( SIGNED_NAT ) -&gt; SIGNED_NAT
236 
</PRE>
237 
which map a literal value to the representation of an integral type.
238 
The token used depends on the list of possible types, which in turn
239 
depends on the base used to represent the literal and the integer
240 
suffix used, as given in the following table:
241 
</P>
242 
<CENTER>
243 
<TABLE BORDER>
244 
<TR><TH>Base</TH>
245 
<TH>Suffix</TH>
246 
<TH>Token</TH>
247 
<TH>Types</TH>
248 
<TR><TD ALIGN=CENTER>decimal</TD>
249 
<TD ALIGN=CENTER>none</TD>
250 
<TD ALIGN=CENTER>~lit_int</TD>
251 
<TD ALIGN=CENTER>int, long, unsigned long</TD>
252 
<TR><TD ALIGN=CENTER>octal</TD>
253 
<TD ALIGN=CENTER>none</TD>
254 
<TD ALIGN=CENTER>~lit_hex</TD>
255 
<TD ALIGN=CENTER>int, unsigned int, long, unsigned long</TD>
256 
<TR><TD ALIGN=CENTER>hexadecimal</TD>
257 
<TD ALIGN=CENTER>none</TD>
258 
<TD ALIGN=CENTER>~lit_hex</TD>
259 
<TD ALIGN=CENTER>int, unsigned int, long, unsigned long</TD>
260 
<TR><TD ALIGN=CENTER>any</TD>
261 
<TD ALIGN=CENTER>U</TD>
262 
<TD ALIGN=CENTER>~lit_unsigned</TD>
263 
<TD ALIGN=CENTER>unsigned int, unsigned long</TD>
264 
<TR><TD ALIGN=CENTER>any</TD>
265 
<TD ALIGN=CENTER>L</TD>
266 
<TD ALIGN=CENTER>~lit_long</TD>
267 
<TD ALIGN=CENTER>long, unsigned long</TD>
268 
<TR><TD ALIGN=CENTER>any</TD>
269 
<TD ALIGN=CENTER>UL</TD>
270 
<TD ALIGN=CENTER>~lit_ulong</TD>
271 
<TD ALIGN=CENTER>unsigned long</TD>
272 
<TR><TD ALIGN=CENTER>any</TD>
273 
<TD ALIGN=CENTER>LL</TD>
274 
<TD ALIGN=CENTER>~lit_longlong</TD>
275 
<TD ALIGN=CENTER>long long, unsigned long long</TD>
276 
<TR><TD ALIGN=CENTER>any</TD>
277 
<TD ALIGN=CENTER>ULL</TD>
278 
<TD ALIGN=CENTER>~lit_ulonglong</TD>
279 
<TD ALIGN=CENTER>unsigned long long</TD>
280 
</TABLE>
281 
</CENTER>
282 
<P>
283 
Thus, for example, the shape of the integer literal 50000 is:
284 
<PRE>
285 
	~convert ( ~lit_int ( 50000 ) )
286 
</PRE>
287 
</P>
288 
 
289 
<HR>
290 
<H3><A NAME="bitfield">2.6.3. Bitfield types</A></H3>
291 
<P>
292 
The sign of a plain bitfield type, declared without using
293 
<CODE>signed</CODE> or <CODE>unsigned</CODE>, is left unspecified
294 
in C and C++.  The token:
295 
<PRE>
296 
	~cpp.bitf_sign : ( SIGNED_NAT ) -&gt; BOOL
297 
</PRE>
298 
is used to give a mapping from integral types to the sign of a plain
299 
bitfield of that type, in a form suitable for use in the TDF
300 
<CODE>bfvar_bits</CODE> construct.  (Note that <CODE>~cpp.bitf_sign</CODE>
301 
should have been a standard C token but was omitted.)
302 
</P>
303 
 
304 
<HR>
305 
<H3><A NAME="pointer">2.6.4. Generic pointers</A></H3>
306 
<P>
307 
TDF has no concept of a generic pointer type, so tokens are used to
308 
defer the representation of <CODE>void *</CODE> and the basic operations
309 
on it to the target machine.  The fundamental token is:
310 
<PRE>
311 
	~ptr_void : () -&gt; SHAPE
312 
</PRE>
313 
which gives the representation of <CODE>void *</CODE>.  This shape
314 
will be denoted by <CODE>pv</CODE> in the description of the following
315 
tokens.  It is not guaranteed that <CODE>pv</CODE> is a TDF <CODE>pointer</CODE>
316 
shape, although normally it will be implemented as a pointer to a
317 
suitable alignment.
318 
</P>
319 
<P>
320 
The token:
321 
<PRE>
322 
	~null_pv : () -&gt; EXP pv
323 
</PRE>
324 
gives the value of a null pointer of type <CODE>void *</CODE>.  Generic
325 
pointers can also be converted to and from other pointers.  These
326 
conversions are represented by the tokens:
327 
<PRE>
328 
	~to_ptr_void : ( ALIGNMENT a, EXP POINTER a ) -&gt; EXP pv
329 
	~from_ptr_void : ( ALIGNMENT a, EXP pv ) -&gt; EXP POINTER a
330 
</PRE>
331 
where the given alignment describes the destination or source pointer
332 
type.  Finally a generic pointer may be tested against the null pointer
333 
or two generic pointers may be compared.  These operations are represented
334 
by the tokens:
335 
<PRE>
336 
	~pv_test : ( EXP pv, LABEL, NTEST ) -&gt; EXP TOP
337 
	~cpp.pv_compare : ( EXP pv, EXP pv, LABEL, NTEST ) -&gt; EXP TOP
338 
</PRE>
339 
where the given <CODE>NTEST</CODE> gives the comparison to be applied
340 
and the given label gives the destination to jump to if the test fails.
341 
(Note that <CODE>~cpp.pv_compare</CODE> should have been a standard
342 
C token but was omitted.)
343 
</P>
344 
 
345 
<HR>
346 
<H3><A NAME="conv">2.6.5. Undefined conversions</A></H3>
347 
<P>
348 
Several conversions in C and C++ can only be represented by undefined
349 
TDF.  For example, converting a pointer to an integer can only be
350 
represented in TDF by forming a union of the pointer and integer shapes,
351 
putting the pointer into the union and pulling the integer out.  Such
352 
conversions are tokenised.  Undefined conversions not mentioned below
353 
may be performed by combining those given with the standard, well-defined,
354 
conversions.
355 
</P>
356 
<P>
357 
The token:
358 
<PRE>
359 
	~ptr_to_ptr : ( ALIGNMENT a, ALIGNMENT b, EXP POINTER a ) -&gt; EXP POINTER b
360 
</PRE>
361 
is used to convert between two incompatible pointer types.  The first
362 
alignment describes the source pointer shape while the second describes
363 
the destination pointer shape.  Note that if the destination alignment
364 
is greater than the source alignment then the source pointer can be
365 
used in most TDF constructs in place of the destination pointer, so
366 
the use of <CODE>~ptr_to_ptr</CODE> can be omitted (the exception
367 
is
368 
<CODE>pointer_test</CODE> which requires equal alignments).  Base
369 
class pointer conversions are examples of these well-behaved, alignment
370 
preserving conversions.
371 
</P>
372 
<P>
373 
The tokens:
374 
<PRE>
375 
	~f_to_pv : ( EXP PROC ) -&gt; EXP pv
376 
	~pv_to_f : ( EXP pv ) -&gt; EXP PROC
377 
</PRE>
378 
are used to convert pointers to functions to and from <CODE>void *</CODE>
379 
(these conversions are not allowed in ISO C/C++ but are in older dialects).
380 
</P>
381 
<P>
382 
The tokens:
383 
<PRE>
384 
	~i_to_p : ( VARIETY v, ALIGNMENT a, EXP INTEGER v ) -&gt; EXP POINTER a
385 
	~p_to_i : ( ALIGNMENT a, VARIETY v, EXP POINTER a ) -&gt; EXP INTEGER v
386 
	~i_to_pv : ( VARIETY v, EXP INTEGER v ) -&gt; EXP pv
387 
	~pv_to_i : ( VARIETY v, EXP pv ) -&gt; EXP INTEGER v
388 
</PRE>
389 
are used to convert integers to and from <CODE>void *</CODE> and other
390 
pointers.
391 
</P>
392 
 
393 
<HR>
394 
<H3><A NAME="div">2.6.6. Integer division</A></H3>
395 
<P>
396 
The precise form of the integer division and remainder operations
397 
in C and C++ is left unspecified with respect to the sign of the result
398 
if either operand is negative.  The tokens:
399 
<PRE>
400 
	~div : ( EXP INTEGER v, EXP INTEGER v ) -&gt; EXP INTEGER v
401 
	~rem : ( EXP INTEGER v, EXP INTEGER v ) -&gt; EXP INTEGER v
402 
</PRE>
403 
are used to represent integer division and remainder.  They will map
404 
onto one of the pairs of TDF constructs, <CODE>div0</CODE> and <CODE>rem0</CODE>,
405 
<CODE>div1</CODE> and <CODE>rem1</CODE> or <CODE>div2</CODE> and
406 
<CODE>rem2</CODE>.
407 
</P>
408 
 
409 
<HR>
410 
<H3><A NAME="call">2.6.7. Calling conventions</A></H3>
411 
<P>
412 
The function calling conventions used by the C++ producer are essentially
413 
the same as those used by the C producer with one exception.  That
414 
is to say, all types except arrays are passed by value (note that
415 
individual installers may modify these conventions to conform to their
416 
own ABIs).
417 
</P>
418 
<P>
419 
The exception concerns classes with a non-trivial constructor, destructor
420 
or assignment operator.  These classes are passed as function arguments
421 
by taking a reference to a copy of the object (although it is often
422 
possible to eliminate the copy and pass a reference to the object
423 
directly).  They are passed as function return values by adding an
424 
extra parameter to the start of the function parameters giving a reference
425 
to a location into which the return value should be copied.
426 
</P>
427 
 
428 
<H4>Member functions</H4>
429 
<P>
430 
Non-static member functions are implemented in the obvious fashion,
431 
by passing a pointer to the object the method is being applied to
432 
as the first argument (or the second argument if the method has an
433 
extra argument for its return value).
434 
</P>
435 
 
436 
<H4><A NAME="ellipsis">Ellipsis functions</A></H4>
437 
<P>
438 
Calls to functions declared with ellipses are via the
439 
<CODE>apply_proc</CODE> TDF construct, with all the arguments being
440 
treated as non-variable.  However the definition of such a function
441 
uses the <CODE>make_proc</CODE> construct with a variable parameter.
442 
This parameter can be referred to within the program using the
443 
<A HREF="pragma.html#ellipsis"><CODE>...</CODE> expression</A>.  The
444 
type of this expression is given by the built-in token:
445 
<PRE>
446 
	~__va_t : () -&gt; SHAPE
447 
</PRE>
448 
The <CODE>va_start</CODE> macro declared in the
449 
<CODE>&lt;stdarg.h&gt;</CODE> header then describes how the variable
450 
parameter (expressed as <CODE>...</CODE>) can be converted to an expression
451 
of type <CODE>va_list</CODE> suitable for use in the
452 
<CODE>va_arg</CODE> macro.
453 
</P>
454 
<P>
455 
Note that the variable parameter is in effect only being used to determine
456 
where the first optional parameter is defined.  The assumption is
457 
that all such parameters are located contiguously on the stack, however
458 
the fact that calls to such functions do not use the variable parameter
459 
mechanism means that this is not automatically the case.  Strictly
460 
speaking this means that the implementation of ellipsis functions
461 
uses undefined behaviour in TDF, however given the non-type-safe function
462 
calling rules in C this is unavoidable and installers need to make
463 
provision for such calls (by dumping any parameters from registers
464 
to the stack if necessary).  Given the theoretically type-safe nature
465 
of C++ it would be possible to avoid such undefined behaviour, but
466 
the need for C-compatible calling conventions prevents this.
467 
</P>
468 
 
469 
<HR>
470 
<H3><A NAME="ptr_mem">2.6.8. Pointers to data members</A></H3>
471 
<P>
472 
The representation of, and operations on, pointers to data members
473 
are represented by tokens to allow for a variety of implementations.
474 
It is assumed that all pointers to data members (as opposed to pointers
475 
to function members) are represented by the same shape:
476 
<PRE>
477 
	~cpp.pm.type : () -&gt; SHAPE
478 
</PRE>
479 
This shape will be denoted by <CODE>pm</CODE> in the description of
480 
the following tokens.
481 
</P>
482 
<P>
483 
There are two basic methods of constructing a pointer to a data member.
484 
The first is to take the address of a data member of a class.  A data
485 
member is represented in TDF by an expression which gives the offset
486 
of the member from the start of its enclosing <CODE>compound</CODE>
487 
shape (note that it is not possible to take the address of a member
488 
of a virtual base). The mapping from this offset to a pointer to a
489 
data member is given by:
490 
<PRE>
491 
	~cpp.pm.make : ( EXP OFFSET ) -&gt; EXP pm
492 
</PRE>
493 
The second way of constructing a pointer to a data member is to use
494 
a null pointer to member:
495 
<PRE>
496 
	~cpp.pm.null : () -&gt; EXP pm
497 
</PRE>
498 
The other fundamental operation on a pointer to data member is to
499 
turn it back into an offset expression which can be added to a pointer
500 
to a class to access a member of that class in a <CODE>.*</CODE> or
501 
<CODE>-&gt*</CODE>
502 
operation.  This is done by the token:
503 
<PRE>
504 
	~cpp.pm.offset : ( EXP pm, ALIGNMENT a ) -&gt; EXP OFFSET ( a, a )
505 
</PRE>
506 
Note that it is necessary to specify an alignment in order to describe
507 
the shape of the result.  The value of this token is undefined if
508 
the given expression is a null pointer to data member.
509 
</P>
510 
<P>
511 
A pointer to a data member of a non-virtual base class can be converted
512 
to a pointer to a data member of a derived class.  The reverse conversion
513 
is also possible using <CODE>static_cast</CODE>.  If the base is a
514 
<A HREF="#primary">primary base class</A> then these conversions are
515 
trivial and have no effect.  Otherwise null pointers to data members
516 
are converted to null pointers to data members, and the non-null cases
517 
are handled by the tokens:
518 
<PRE>
519 
	~cpp.pm.cast : ( EXP pm, EXP OFFSET ) -&gt; EXP pm
520 
	~cpp.pm.uncast : ( EXP pm, EXP OFFSET ) -&gt; EXP pm
521 
</PRE>
522 
where the given offset is the offset of the base class within the
523 
derived class.  It is also possible to convert between any two pointers
524 
to data members using <CODE>reinterpret_cast</CODE>.  This conversion
525 
is implied by the equality of representation between any two pointers
526 
to data members and has no effect.
527 
</P>
528 
<P>
529 
The only remaining operations on pointer to data members are to test
530 
one against the null pointer to data member and to compare two pointer
531 
to data members.  These are represented by the tokens:
532 
<PRE>
533 
	~cpp.pm.test : ( EXP pm, LABEL, NTEST ) -&gt; EXP TOP
534 
	~cpp.pm.compare : ( EXP pm, EXP pm, LABEL, NTEST ) -&gt; EXP TOP
535 
</PRE>
536 
where the given <CODE>NTEST</CODE> gives the comparison to be applied
537 
and the given label gives the destination to jump to if the test fails.
538 
</P>
539 
<P>
540 
In the default implementation, pointers to data members are implemented
541 
as <CODE>int</CODE>.  The null pointer to member is represented by
542 
 
543 
of the member (in bytes).  Casting to and from a derived class then
544 
correspond to adding or subtracting the base class offset (in bytes),
545 
and pointer to member comparisons correspond to integer comparisons.
546 
</P>
547 
 
548 
<HR>
549 
<H3><A NAME="ptr_mem_func">2.6.9. Pointers to function members</A></H3>
550 
<P>
551 
As with pointers to data members, pointers to function members and
552 
the operations on them are represented by tokens to allow for a range
553 
of implementations.  All pointers to function members are represented
554 
by the same shape:
555 
<PRE>
556 
	~cpp.pmf.type : () -&gt; SHAPE
557 
</PRE>
558 
This shape will be denoted by <CODE>pmf</CODE> in the description
559 
of the following tokens.  Many of the tokens take an expression which
560 
has a shape which is a pointer to the alignment of <CODE>pmf</CODE>.
561 
This will be denoted by <CODE>ppmf</CODE>.
562 
</P>
563 
<P>
564 
There are two basic methods for constructing a pointer to a function
565 
member.  The first is to take the address of a non-static member function
566 
of a class.  There are two cases, depending on whether or not the
567 
member function is virtual.  The non-virtual case is given by the
568 
token:
569 
<PRE>
570 
	~cpp.pmf.make : ( EXP PROC, EXP OFFSET, EXP OFFSET ) -&gt; EXP pmf
571 
</PRE>
572 
where the first argument is the address of the corresponding function,
573 
the second argument gives any base class offset which is to be added
574 
when calling this function (to deal with inherited member functions),
575 
and the third argument is a zero offset.
576 
</P>
577 
<P>
578 
For virtual functions, a pointer to function member of the form above
579 
is entered in the <A HREF="#vtable">virtual function table</A> for
580 
the corresponding class.  The actual pointer to the virtual function
581 
member then gives a reference into the virtual function table as follows:
582 
<PRE>
583 
	~cpp.pmf.vmake : ( SIGNED_NAT, EXP OFFSET, EXP, EXP ) -&gt; EXP pmf
584 
</PRE>
585 
where the first argument gives the index of the function within the
586 
virtual function table, the second argument gives the offset of the
587 
<I>vptr</I> field within the class, and the third and fourth arguments
588 
are zero offsets.
589 
</P>
590 
<P>
591 
The second way of constructing a pointer to a function member is to
592 
use a null pointer to function member:
593 
<PRE>
594 
	~cpp.pmf.null : () -&gt; EXP pmf
595 
	~cpp.pmf.null2 : () -&gt; EXP pmf
596 
</PRE>
597 
For technical reasons there are two versions of this token, although
598 
they have the same value.  The first token is used in static initialisers;
599 
the second token is used in other expressions.
600 
<P>
601 
The cast operations on pointers to function members are more complex
602 
than those on pointers to data members.  The value to be cast is copied
603 
into a temporary and one of the tokens:
604 
<PRE>
605 
	~cpp.pmf.cast : ( EXP ppmf, EXP OFFSET, EXP, EXP OFFSET ) -&gt; EXP TOP
606 
	~cpp.pmf.uncast : ( EXP ppmf, EXP OFFSET, EXP, EXP OFFSET ) -&gt; EXP TOP
607 
</PRE>
608 
is applied to modify the value of the temporary according to the given
609 
cast.  The first argument gives the address of the temporary, the
610 
second gives the base class offset to be added or subtracted, the
611 
third gives the number to be added or subtracted to convert virtual
612 
function indexes for the base class into virtual function indexes
613 
for the derived class, and the fourth gives the offset of the <I>vptr</I>
614 
field within the class.  Again, the ability to use <CODE>reinterpret_cast</CODE>
615 
to convert between any two pointer to function member types arises
616 
because of the uniform representation of these types.
617 
</P>
618 
<P>
619 
As with pointers to data members, there are tokens implementing comparisons
620 
on pointers to function members:
621 
<PRE>
622 
	~cpp.pmf.test : ( EXP ppmf, LABEL, NTEST ) -&gt; EXP TOP
623 
	~cpp.pmf.compare : ( EXP ppmf, EXP ppmf, LABEL, NTEST ) -&gt; EXP TOP
624 
</PRE>
625 
Note however that the arguments are passed by reference.
626 
</P>
627 
<P>
628 
The most important, and most complex, operation is calling a function
629 
through a pointer to function member.  The first step is to copy the
630 
pointer to function member into a temporary.  The token:
631 
<PRE>
632 
	~cpp.pmf.virt : ( EXP ppmf, EXP, ALIGNMENT ) -&gt; EXP TOP
633 
</PRE>
634 
is then applied to the temporary to convert a pointer to a virtual
635 
function member to a normal pointer to function member by looking
636 
it up in the corresponding virtual function table.  The first argument
637 
gives the address of the temporary, the second gives the object to
638 
which the function is to be applied, and the third gives the alignment
639 
of the corresponding class.  Now the base class conversion to be applied
640 
to the object can be determined by applying the token:
641 
<PRE>
642 
	~cpp.pmf.delta : ( ALIGNMENT a, EXP ppmf ) -&gt; EXP OFFSET ( a, a )
643 
</PRE>
644 
to the temporary to find the offset to be added.  Finally the function
645 
to be called can be extracted from the temporary using the token:
646 
<PRE>
647 
	~cpp.pmf.func : ( EXP ppmf ) -&gt; EXP PROC
648 
</PRE>
649 
The function call then procedes as normal.
650 
</P>
651 
<P>
652 
The default implementation is that described in the ARM, where each
653 
pointer to function member is represented in the form:
654 
<PRE>
655 
	struct PTR_MEM_FUNC {
656 
	    short delta ;
657 
	    short index ;
658 
	    union {
659 
		void ( *func ) () ;
660 
		short off ;
661 
	    } u ;
662 
	} ;
663 
</PRE>
664 
The <CODE>delta</CODE> field gives the base class offset (in bytes)
665 
to be added before applying the function.  The <CODE>index</CODE>
666 
field is 0 for null pointers, -1 for non-virtual function pointers
667 
and the index into the virtual function table for virtual function
668 
pointers (as described below these indexes start from 1).  For non-virtual
669 
function pointers the function itself is given by the <CODE>u.func</CODE>
670 
field. For virtual function pointers the offset of the <I>vptr</I>
671 
field within the class is given by the <CODE>u.off</CODE> field.
672 
</P>
673 
 
674 
<HR>
675 
<H3><A NAME="class">2.6.10. Class layout</A></H3>
676 
<P>
677 
Consider a class with no base classes:
678 
<PRE>
679 
	class A {
680 
	    // A's members
681 
	} ;
682 
</PRE>
683 
Each object of class <I>A</I> needs its own copy of the non-static
684 
data members of <I>A</I> and, for polymorphic types, a means of referencing
685 
the virtual function table and run-time type information for <I>A</I>.
686 
This is accomplished using a layout of the form:
687 
<CENTER>
688 
<IMG SRC="../images/class.gif" ALT="class A">
689 
</CENTER>
690 
where the <I>A</I> component consists of the non-static data members
691 
and
692 
<I>vptr A</I> is a pointer to the virtual function table for <I>A</I>.
693 
For non-polymorphic classes the <I>vptr A</I> field is omitted; otherwise
694 
space for <I>vptr A</I> needs to be allocated within the class and
695 
the pointer needs to be initialised in each constructor for <I>A</I>.
696 
The precise layout of the <A HREF="#vtable">virtual function table</A>
697 
and the <A HREF="#rtti">run-time type information</A> is given below.
698 
</P>
699 
<P>
700 
Two alternative ways of laying out the non-static data members within
701 
the class are implemented.  The first, which is default, gives them
702 
in the order in which they are declared in the class definition.
703 
The second lays out the <CODE>public</CODE>, the <CODE>protected</CODE>,
704 
and the <CODE>private</CODE> members in three distinct sections, the
705 
members within each section being given in the order in which they
706 
are declared. The latter can be enabled using the <CODE>-jo</CODE>
707 
command-line option.
708 
</P>
709 
<P>
710 
The offset of each member within the class (including <I>vptr A</I>)
711 
can be calculated in terms of the offset of the previous member.
712 
The first member has offset zero.  The offset of any other member
713 
is given by the offset of the previous member plus the size of the
714 
previous member, rounded up to the alignment of the current member.
715 
The overall size of the class is given by the offset of the last member
716 
plus the size of the last member, rounded up using the token:
717 
<PRE>
718 
	~comp_off : ( EXP OFFSET ) -&gt; EXP OFFSET
719 
</PRE>
720 
which allows for any target dependent padding at the end of the class.
721 
The shape of the class is then a <CODE>compound</CODE> shape with
722 
this offset.
723 
</P>
724 
<P>
725 
Classes with no members need to be treated slightly differently.
726 
The shape of such a class is given by the token:
727 
<PRE>
728 
	~cpp.empty.shape : () -&gt; SHAPE
729 
</PRE>
730 
(recall that an empty class still has a nonzero size).  The token:
731 
<PRE>
732 
	~cpp.empty.offset : () -&gt; EXP OFFSET
733 
</PRE>
734 
is used to represent the offset required for an empty class when it
735 
is used as a base class.  This may be a zero offset.
736 
</P>
737 
<P>
738 
Bitfield members provide a slight complication to the picture above.
739 
The offset of a bitfield is additionally padded using the token:
740 
<PRE>
741 
	~pad : ( EXP OFFSET, SHAPE, SHAPE ) -&gt; EXP OFFSET
742 
</PRE>
743 
where the two shapes give the type underlying the bitfield and the
744 
bitfield itself.
745 
</P>
746 
<P>
747 
The layout of unions is similar to that of classes except that all
748 
members have zero offset, and the size of the union is the maximum
749 
of the sizes of its members, suitably padded.  Of course unions cannot
750 
be polymorphic and cannot have base classes.
751 
</P>
752 
<P>
753 
Pointers to incomplete classes are represented by means of the alignment:
754 
<PRE>
755 
	~cpp.empty.align : () -&gt; ALIGNMENT
756 
</PRE>
757 
This token is also used for the alignment of a complete class if that
758 
class is never used in the generated TDF in a manner which requires
759 
it to be complete.  This can lead to savings on the size of the generated
760 
code by preventing the need to define all the member offset tokens
761 
in order to find the shape of the class.
762 
</P>
763 
 
764 
<HR>
765 
<H3><A NAME="derive">2.6.11. Derived class layout</A></H3>
766 
<P>
767 
The description of the implementation of derived classes will be given
768 
in terms of the example class hierarchy given by:
769 
<PRE>
770 
	class A {
771 
	    // A's members
772 
	} ;
773 
 
774 
	class B : public A {
775 
	    // B's members
776 
	} ;
777 
 
778 
	class C : public A {
779 
	    // C's members
780 
	} ;
781 
 
782 
	class D : public B, public C {
783 
	    // D's members
784 
	} ;
785 
</PRE>
786 
or, as a directed acyclic graph:
787 
<CENTER>
788 
<IMG SRC="../images/graph.gif" ALT="class D">
789 
</CENTER>
790 
 
791 
<H4>Single inheritance</H4>
792 
<P>
793 
The layout of class <I>A</I> is given by:
794 
<CENTER>
795 
<IMG SRC="../images/classA.gif" ALT="class A">
796 
</CENTER>
797 
as above.  Class <I>B</I> inherits all the members of class <I>A</I>
798 
plus those members explicitly declared within class <I>B</I>.  In
799 
addition, class <I>B</I> inherits all the virtual member functions
800 
of <I>A</I>, some of which may be overridden in <I>B</I>, extended
801 
by any additional virtual functions declared in <I>B</I>.  This may
802 
be represented as follows:
803 
<CENTER>
804 
<IMG SRC="../images/classB.gif" ALT="class B">
805 
</CENTER>
806 
where <I>A</I> denotes those members inherited from the base class
807 
and
808 
<I>B</I> denotes those members added in the derived class.  Note that
809 
an object of class <I>B</I> contains a sub-object of class <I>A</I>.
810 
The fact that this sub-object is located at the start of <I>B</I>
811 
means that the base class conversion from <I>B</I> to <I>A</I> is
812 
trivial.  Any base class with this property is called a
813 
<A NAME="primary">primary base class</A>.
814 
</P>
815 
<P>
816 
Note that in theory two virtual function tables are required, the
817 
normal virtual function table for <I>B</I>, denoted by <I>vtbl B</I>,
818 
and a modified virtual function table for <I>A</I>, denoted by <I>vtbl
819 
B::A</I>, taking into account any overriding virtual functions within
820 
<I>B</I>, and pointing to <I>B</I>'s run-time type information.  This
821 
latter means that the dynamic type information for the <I>A</I> sub-object
822 
relates to
823 
<I>B</I> rather than <I>A</I>.  However these two tables can usually
824 
be combined - if the virtual functions added in <I>B</I> are listed
825 
in the virtual function table after those inherited from <I>A</I>
826 
and the form of the overriding is <A HREF="#override">suitably well
827 
behaved</A>
828 
(in the sense defined below) then <I>vptr B::A</I> is an initial segment
829 
of <I>vptr B</I>.  It is also possible to remove the <I>vptr B</I>
830 
field and use <I>vptr B::A</I> in its place in this case (it has to
831 
be this way round to preserve the <I>A</I> sub-object).  Thus the
832 
items shaded in the diagram can be removed.
833 
</P>
834 
<P>
835 
The class <I>C</I> is similarly given by:
836 
<CENTER>
837 
<IMG SRC="../images/classC.gif" ALT="class C">
838 
</CENTER>
839 
</P>
840 
 
841 
<H4>Multiple inheritance</H4>
842 
<P>
843 
Class <I>D</I> is more complex because of the presence of multiple
844 
inheritance.  <I>D</I> inherits all the members of <I>B</I>, including
845 
those which <I>B</I> inherits from <I>A</I>, plus all the members
846 
of
847 
<I>C</I>, including those which <I>C</I> inherits from <I>A</I>.
848 
It also inherits all of the virtual member functions from <I>B</I>
849 
and
850 
<I>C</I>, some of which may be overridden in <I>D</I>, extended by
851 
any additional virtual functions declared in <I>D</I>.  This may be
852 
represented as follows:
853 
<CENTER>
854 
<IMG SRC="../images/classD.gif" ALT="class D">
855 
</CENTER>
856 
Note that there are two copies of <I>A</I> in <I>D</I> because virtual
857 
inheritance has not been used.
858 
</P>
859 
<P>
860 
The <I>B</I> base class of <I>D</I> is essentially similar to the
861 
single inheritance case already discussed; the <I>C</I> base class
862 
is different however.  Note firstly that the <I>C</I> sub-object of
863 
<I>D</I> is located at a non-zero offset, <I>delta D::C</I>, from
864 
the start of the object. This means that the base class conversion
865 
from <I>D</I> to <I>C</I>
866 
consists of adding this offset (for pointer conversions things are
867 
further complicated by the need to allow for null pointers).  Also
868 
<I>vtbl D::C</I> is not an initial segment of <I>vtbl D</I> because
869 
this contains the virtual functions inherited from <I>B</I> first,
870 
followed by those inherited from <I>C</I>, followed by those first
871 
declared in <I>D</I> (there are <A HREF="#override">other reasons</A>
872 
as well).  Thus <I>vtbl D::C</I> cannot be eliminated.
873 
</P>
874 
 
875 
<H4>Virtual inheritance</H4>
876 
<P>
877 
Virtual inheritance introduces a further complication.  Now consider
878 
the class hierarchy given by:
879 
<PRE>
880 
	class A {
881 
	    // A's members
882 
	} ;
883 
 
884 
	class B : virtual public A {
885 
	    // B's members
886 
	} ;
887 
 
888 
	class C : virtual public A {
889 
	    // C's members
890 
	} ;
891 
 
892 
	class D : public B, public C {
893 
	    // D's members
894 
	} ;
895 
</PRE>
896 
or, as a <A NAME="diamond">directed acyclic graph</A>:
897 
<CENTER>
898 
<IMG SRC="../images/diamond.gif" ALT="class D">
899 
</CENTER>
900 
As before <I>A</I> is given by:
901 
<CENTER>
902 
<IMG SRC="../images/classA.gif" ALT="class A">
903 
</CENTER>
904 
but now <I>B</I> is given by:
905 
<CENTER>
906 
<IMG SRC="../images/virtualB.gif" ALT="class B">
907 
</CENTER>
908 
Rather than having the sub-object of class <I>A</I> directly as part
909 
of
910 
<I>B</I>, the class now contains a pointer, <I>ptr A</I>, to this
911 
sub-object.  The virtual sub-objects are always located at the end
912 
of a class layout; their offset may therefore vary for different objects,
913 
however the offset for <I>ptr A</I> is always fixed.  The <I>ptr A</I>
914 
field is initialised in each constructor for <I>B</I>.  In order to
915 
perform the base class conversion from <I>B</I> to <I>A</I>, the contents
916 
of <I>ptr A</I> are taken (again provision needs to be made for null
917 
pointers in pointer conversions).  In cases when the dynamic type
918 
of the <I>B</I> object can be determined statically it is possible
919 
to access the <I>A</I> sub-object directly by adding a suitable offset.
920 
Because this conversion is non-trivial (see <A HREF="#override">below</A>)
921 
the virtual function table <I>vtbl B::A</I> is not an initial segment
922 
of
923 
<I>vtbl B</I> and cannot be eliminated.
924 
</P>
925 
<P>
926 
The class <I>C</I> is similarly given by:
927 
<CENTER>
928 
<IMG SRC="../images/virtualC.gif" ALT="class C">
929 
</CENTER>
930 
Now the class <I>D</I> is given by:
931 
<CENTER>
932 
<IMG SRC="../images/virtualD.gif" ALT="class D">
933 
</CENTER>
934 
Note that there is a single <I>A</I> sub-object of <I>D</I> referenced
935 
by the <I>ptr A</I> fields in both the <I>B</I> and <I>C</I> sub-objects.
936 
The elimination of <I>vtbl D::B</I> is as above.
937 
</P>
938 
 
939 
<HR>
940 
<H3><A NAME="constr">2.6.12. Constructors and destructors</A></H3>
941 
<P>
942 
The implementation of constructors and destructors, whether explicitly
943 
or implicitly defined, is slightly more complex than that of other
944 
member functions.  For example, the constructors need to set up the
945 
internal <I>vptr</I> and <I>ptr</I> fields mentioned above.
946 
</P>
947 
<P>
948 
The order of initialisation in a constructor is as follows:
949 
<OL>
950 
<LI>The internal <I>ptr</I> fields giving the locations of the virtual
951 
base classes are initialised.
952 
<LI>The constructors for the virtual base classes are called.
953 
<LI>The constructors for the non-virtual direct base classes are called.
954 
<LI>The internal <I>vptr</I> fields giving the locations of the virtual
955 
function tables are initialised.
956 
<LI>The constructors for the members of the class are called.
957 
<LI>The main constructor body is executed.
958 
</OL>
959 
To ensure that each virtual base is only initialised once, if a class
960 
has a virtual base class then all its constructors have an implicit
961 
extra parameter of type <CODE>int</CODE>.  The first two steps above
962 
are then only applied if this flag is nonzero.  In normal applications
963 
of the constructor this argument will be 1, however in base class
964 
initialisations such as those in the third and fourth steps above,
965 
it will be 0.
966 
</P>
967 
<P>
968 
Note that similar steps to protect virtual base classes are not taken
969 
in an implicitly declared <CODE>operator=</CODE> function.  The order
970 
of assignment in this case is as follows:
971 
<OL>
972 
<LI>The assignment operators for the direct base classes (both virtual
973 
and non-virtual) are called.
974 
<LI>The assignment operators for the members of the class are called.
975 
<LI>A reference to the object assigned to (i.e. <CODE>*this</CODE>)
976 
is     returned.
977 
</OL>
978 
</P>
979 
<P>
980 
The order of destruction in a destructor is essentially the reverse
981 
of the order of construction:
982 
<OL>
983 
<LI>The main destructor body is executed.
984 
<LI>The destructor for the members of the class are called.
985 
<LI>The internal <I>vptr</I> fields giving the locations of the virtual
986 
function tables are re-initialised.
987 
<LI>The destructors for the non-virtual direct base classes are called.
988 
<LI>The destructors for the virtual base classes are called.
989 
<LI>If necessary the space occupied by the object is deallocated.
990 
</OL>
991 
All destructors have an extra parameter of type <CODE>int</CODE>.
992 
The virtual base classes are only destroyed if this flag is nonzero
993 
when and-ed with 2.  The space occupied by the object is only deallocated
994 
if this flag is nonzero when and-ed with 1.  This deallocation is
995 
equivalent to inserting:
996 
<PRE>
997 
	delete this ;
998 
</PRE>
999 
in the destructor.  The <CODE>operator delete</CODE> function is called
1000 
via the destructor in this way in order to implement the pseudo-virtual
1001 
nature of these deallocation functions.  Thus for normal destructor
1002 
calls the extra argument is 2, for base class destructor calls it
1003 
is 0, and for calls arising from a <CODE>delete</CODE> expression
1004 
it is 3.
1005 
</P>
1006 
<P>
1007 
The point at which the virtual function tables are initialised in
1008 
the constructor, and the fact that they are re-initialised in the
1009 
destructor, is to ensure that virtual functions called from base class
1010 
initialisers are handled correctly (see ISO C++ 12.7).
1011 
</P>
1012 
<P>
1013 
A further complication arises from the need to destroy
1014 
<A NAME="partial">partially constructed objects</A> if an exception
1015 
is thrown in a constructor.  A count is maintained of the number of
1016 
base classes and members constructed within a constructor.  If an
1017 
exception is thrown then it is caught in the constructor, the constructed
1018 
base classes and members are destroyed, and the exception is re-thrown.
1019 
The count variable is used to determine which bases and members need
1020 
to be destroyed.
1021 
</P>
1022 
<P>
1023 
<IMG SRC="../images/warn.gif" ALT="warning"> These partial destructors
1024 
currently do not interact correctly with any exception specification
1025 
on the constructor.  Exceptions thrown within destructors are not
1026 
correctly handled either.
1027 
</P>
1028 
 
1029 
<HR>
1030 
<H3><A NAME="vtable">2.6.13. Virtual function tables</A></H3>
1031 
<P>
1032 
The virtual functions in a polymorphic class are given in its virtual
1033 
function table in the following order: firstly those virtual functions
1034 
inherited from its direct base classes (which may be overridden in
1035 
the derived class) followed by those first declared in the derived
1036 
class in the order in which they are declared.  Note that this can
1037 
result in virtual functions inherited from virtual base classes appearing
1038 
more than once.  The virtual functions are numbered from 1 (this is
1039 
slightly more convenient than numbering from 0 in the default implementation).
1040 
</P>
1041 
<P>
1042 
The virtual function table for this class has shape:
1043 
<PRE>
1044 
	~cpp.vtab.type : ( NAT ) -&gt; SHAPE
1045 
</PRE>
1046 
the argument being <I>n + 1</I> where <I>n</I> is the number of virtual
1047 
functions in the class (there is also a token:
1048 
<PRE>
1049 
	~cpp.vtab.diag : () -&gt; SHAPE
1050 
</PRE>
1051 
which is used in the diagnostic output for a generic virtual function
1052 
table).  The table is created using the token:
1053 
<PRE>
1054 
	~cpp.vtab.make : ( EXP pti, EXP OFFSET, NAT, EXP NOF ) -&gt; EXP vt
1055 
</PRE>
1056 
where the first expression gives the address of the <A HREF="#rtti">run-time
1057 
type information structure</A> for the class, the second expression
1058 
gives the offset of the <I>vptr</I> field within the class (i.e. <I>voff</I>),
1059 
the integer constant is <I>n + 1</I>, and the final expression is
1060 
a
1061 
<CODE>make_nof</CODE> construct giving information on each of the
1062 
<I>n</I>
1063 
virtual functions.
1064 
</P>
1065 
<P>
1066 
The information given on each virtual function in this table has the
1067 
form of a <A HREF="#ptr_mem_func">pointer to function member</A> formed
1068 
using the token:
1069 
<PRE>
1070 
	~cpp.pmf.make : ( EXP PROC, EXP OFFSET, EXP OFFSET ) -&gt; EXP pmf
1071 
</PRE>
1072 
as above, except that the third argument gives the offset of the base
1073 
class in virtual function tables such as <I>vtbl B::A</I>.  For pure
1074 
virtual functions the function pointer in this token is given by:
1075 
<PRE>
1076 
	~cpp.vtab.pure : () -&gt; EXP PROC
1077 
</PRE>
1078 
In the default implementation this gives a function
1079 
<CODE>__TCPPLUS_pure</CODE> which just calls <CODE>abort</CODE>.
1080 
</P>
1081 
<P>
1082 
To avoid duplicate copies of virtual function tables and run-time
1083 
type information structures being created, the ARM algorithm is used.
1084 
The virtual function table and run-time type information structure
1085 
for a class are defined in the module containing the definition of
1086 
the first non-inline, non-pure virtual function declared in that class.
1087 
If such a function does not exist then duplicate copies are created
1088 
in every module which requires them.  In the former case the virtual
1089 
function table will have an <A HREF="#other">external tag name</A>;
1090 
in the latter case it will be an internal tag.  This scheme can be
1091 
overridden using the <CODE>-jv</CODE> command-line option, which causes
1092 
local virtual function tables to be output for all classes.
1093 
</P>
1094 
<P>
1095 
Note that the discussion above applies to both simple virtual function
1096 
tables, such as <I>vtbl B</I> above, and to those arising from base
1097 
classes, such as <I>vtbl B::A</I>.  <A NAME="override">We are now
1098 
in a position to precisely determine when <I>vtbl B::A</I> is an initial
1099 
segment of <I>vtbl B</I> and hence can be eliminated</A>.  Firstly,
1100 
<I>A</I> must be the first direct base class of <I>B</I> and cannot
1101 
be virtual.  This is to ensure both that there are no virtual functions
1102 
in <I>vtbl B</I> before those inherited from <I>A</I>, and that the
1103 
corresponding base class conversion is trivial so that the pointers
1104 
to function members of <I>B</I> comprising the virtual function table
1105 
can be equally regarded as pointers to function members of <I>A</I>.
1106 
The second requirement is that if a virtual function for <I>A</I>,
1107 
<I>f</I>, is overridden in <I>B</I> then the return type for <I>B::f</I>
1108 
cannot differ from the return type for <I>A::f</I> by a non-trivial
1109 
conversion (recall that ISO C++ allows the return types to differ
1110 
by a base class conversion).  In the non-trivial conversion case the
1111 
function entered in <I>vtbl B::A</I> needs to be, not <I>B::f</I>
1112 
as in <I>vtbl B</I>, but a stub function which calls <I>B::f</I> and
1113 
converts its return value to the return type of <I>A::f</I>.
1114 
</P>
1115 
 
1116 
<H4>Calling virtual functions</H4>
1117 
<P>
1118 
The virtual function call mechanism is implemented using the token:
1119 
<PRE>
1120 
	~cpp.vtab.func : ( EXP ppvt, SIGNED_NAT ) -&gt; EXP ppmf
1121 
</PRE>
1122 
which has as its arguments a reference to the <I>vptr</I> field of
1123 
the object the function is to be called for, and the number of the
1124 
virtual function to be called.  It returns a reference to the corresponding
1125 
pointer to function member within the object's virtual function table.
1126 
The function is then called by extracting the base class offset to
1127 
be added, and the function to be called, from this reference using
1128 
the tokens:
1129 
<PRE>
1130 
	~cpp.pmf.delta : ( ALIGNMENT a, EXP ppmf ) -&gt; EXP OFFSET ( a, a )
1131 
	~cpp.pmf.func : ( EXP ppmf ) -&gt; EXP PROC
1132 
</PRE>
1133 
described as part of the <A HREF="#ptr_mem_func">pointer to function
1134 
member call mechanism</A> above.
1135 
</P>
1136 
 
1137 
<HR>
1138 
<H3><A NAME="rtti">2.6.14. Run-time type information</A></H3>
1139 
<P>
1140 
Each C++ type can be associated with a run-time type information structure
1141 
giving information about that type.  These type information structures
1142 
have shape given by the token:
1143 
<PRE>
1144 
	~cpp.typeid.type : () -&gt; SHAPE
1145 
</PRE>
1146 
which corresponds to the representation for the standard type
1147 
<CODE>std::type_info</CODE> declared in the header
1148 
<CODE>&lt;typeinfo&gt;</CODE>.  Each type information structure consists
1149 
of a tag number, giving information on the kind of type represented,
1150 
a string literal, giving the name of the type, and a pointer to a
1151 
list of base type information structures.  These are combined to give
1152 
a type information structure using the token:
1153 
<PRE>
1154 
	~cpp.typeid.make : ( SIGNED_NAT, EXP, EXP ) -&gt; EXP ti
1155 
</PRE>
1156 
Each base type information structure has shape given by the token:
1157 
<PRE>
1158 
	~cpp.baseid.type : () -&gt; SHAPE
1159 
</PRE>
1160 
It consists of a pointer to a type information structure, an expression
1161 
used to describe the offset of a base class, a pointer to the next
1162 
base type information structure in the list, and two integers giving
1163 
information on type qualifiers etc.  These are combined to give a
1164 
base type information structure using the token:
1165 
<PRE>
1166 
	~cpp.baseid.make : ( EXP, EXP, EXP, SIGNED_NAT, SIGNED_NAT ) -&gt; EXP bi
1167 
</PRE>
1168 
</P>
1169 
<P>
1170 
The following table gives the various tag numbers used in type information
1171 
structures plus a list of the base type information structures associated
1172 
with each type.  Macros giving these tag numbers are provided in the
1173 
default implementation in a header, <CODE>interface.h</CODE>, which
1174 
is shared by the C++ producer.
1175 
</P>
1176 
<P>
1177 
<CENTER>
1178 
<TABLE BORDER>
1179 
<TR><TH>Type</TH>
1180 
<TH>Form</TH>
1181 
<TH>Tag</TH>
1182 
<TH>Base information</TH>
1183 
<TR><TD ALIGN=CENTER>integer</TD>
1184 
<TD ALIGN=CENTER>-</TD>
1185 
<TD ALIGN=CENTER>0</TD>
1186 
<TD ALIGN=CENTER>-</TD>
1187 
<TR><TD ALIGN=CENTER>floating point</TD>
1188 
<TD ALIGN=CENTER>-</TD>
1189 
<TD ALIGN=CENTER>1</TD>
1190 
<TD ALIGN=CENTER>-</TD>
1191 
<TR><TD ALIGN=CENTER>void</TD>
1192 
<TD ALIGN=CENTER>-</TD>
1193 
<TD ALIGN=CENTER>2</TD>
1194 
<TD ALIGN=CENTER>-</TD>
1195 
<TR><TD ALIGN=CENTER>class or struct</TD>
1196 
<TD ALIGN=CENTER>class T</TD>
1197 
<TD ALIGN=CENTER>3</TD>
1198 
<TD ALIGN=CENTER>[base,access,virtual], ....</TD>
1199 
<TR><TD ALIGN=CENTER>union</TD>
1200 
<TD ALIGN=CENTER>union T</TD>
1201 
<TD ALIGN=CENTER>4</TD>
1202 
<TD ALIGN=CENTER>-</TD>
1203 
<TR><TD ALIGN=CENTER>enumeration</TD>
1204 
<TD ALIGN=CENTER>enum T</TD>
1205 
<TD ALIGN=CENTER>5</TD>
1206 
<TD ALIGN=CENTER>-</TD>
1207 
<TR><TD ALIGN=CENTER>pointer</TD>
1208 
<TD ALIGN=CENTER>cv T *</TD>
1209 
<TD ALIGN=CENTER>6</TD>
1210 
<TD ALIGN=CENTER>[T,cv,0]</TD>
1211 
<TR><TD ALIGN=CENTER>reference</TD>
1212 
<TD ALIGN=CENTER>cv T &amp;</TD>
1213 
<TD ALIGN=CENTER>7</TD>
1214 
<TD ALIGN=CENTER>[T,cv,0]</TD>
1215 
<TR><TD ALIGN=CENTER>pointer to member</TD>
1216 
<TD ALIGN=CENTER>cv T S::*</TD>
1217 
<TD ALIGN=CENTER>8</TD>
1218 
<TD ALIGN=CENTER>[S,0,0], [T,cv,0]</TD>
1219 
<TR><TD ALIGN=CENTER>array</TD>
1220 
<TD ALIGN=CENTER>cv T [n]</TD>
1221 
<TD ALIGN=CENTER>9</TD>
1222 
<TD ALIGN=CENTER>[T,cv,n]</TD>
1223 
<TR><TD ALIGN=CENTER>bitfield</TD>
1224 
<TD ALIGN=CENTER>cv T : n</TD>
1225 
<TD ALIGN=CENTER>10</TD>
1226 
<TD ALIGN=CENTER>[T,cv,n]</TD>
1227 
<TR><TD ALIGN=CENTER>C++ function</TD>
1228 
<TD ALIGN=CENTER>cv T ( S1, ...., Sn )</TD>
1229 
<TD ALIGN=CENTER>11</TD>
1230 
<TD ALIGN=CENTER>[T,cv,0], [S1,0,0], ...., [Sn,0,0]</TD>
1231 
<TR><TD ALIGN=CENTER>C function</TD>
1232 
<TD ALIGN=CENTER>cv T ( S1, ...., Sn )</TD>
1233 
<TD ALIGN=CENTER>12</TD>
1234 
<TD ALIGN=CENTER>[T,cv,0], [S1,0,0], ...., [Sn,0,0]</TD>
1235 
</TABLE>
1236 
</CENTER>
1237 
</P>
1238 
<P>
1239 
In the form column <CODE>cv T</CODE> is used to denote not only the
1240 
normal cv-qualifiers but, when <CODE>T</CODE> is a function type,
1241 
the member function cv-qualifiers.  Arrays with an unspecified bound
1242 
are treated as if their bound was zero.  Functions with ellipsis are
1243 
treated as if they had an extra parameter of a dummy type named
1244 
<CODE>...</CODE> (see below).  Note the distinction between C++ and
1245 
C function types.
1246 
</P>
1247 
<P>
1248 
Each base type information structure is described as a triple consisting
1249 
of a type and two integers.  One of these integers may be used to
1250 
encode a type qualifier, <CODE>cv</CODE>, as follows:
1251 
</P>
1252 
<P>
1253 
<CENTER>
1254 
<TABLE BORDER>
1255 
<TR><TH>Qualifier</TH>   <TH>Encoding</TH>
1256 
<TR><TD ALIGN=CENTER>none</TD>  <TD ALIGN=CENTER>0</TD>
1257 
<TR><TD ALIGN=CENTER>const</TD>  <TD ALIGN=CENTER>1</TD>
1258 
<TR><TD ALIGN=CENTER>volatile</TD> <TD ALIGN=CENTER>2</TD>
1259 
<TR><TD ALIGN=CENTER>const volatile</TD><TD ALIGN=CENTER>3</TD>
1260 
</TABLE>
1261 
</CENTER>
1262 
</P>
1263 
<P>
1264 
The base type information for a class consists of information on each
1265 
of its direct base classes.  The includes the offset of this base
1266 
within the class (for a virtual base class this is the offset of the
1267 
corresponding
1268 
<I>ptr</I> field), whether the base is virtual (1) or not (0), and
1269 
the base class access, encoded as follows:
1270 
</P>
1271 
<P>
1272 
<CENTER>
1273 
<TABLE BORDER>
1274 
<TR><TH>Access</TH>   <TH>Encoding</TH>
1275 
<TR><TD ALIGN=CENTER>public</TD> <TD ALIGN=CENTER>0</TD>
1276 
<TR><TD ALIGN=CENTER>protected</TD> <TD ALIGN=CENTER>1</TD>
1277 
<TR><TD ALIGN=CENTER>private</TD> <TD ALIGN=CENTER>2</TD>
1278 
</TABLE>
1279 
</CENTER>
1280 
</P>
1281 
<P>
1282 
For example, the run-time type information structures for the classes
1283 
declared in the <A HREF="#diamond">diamond lattice</A> above can be
1284 
represented as follows:
1285 
<CENTER>
1286 
<IMG SRC="../images/rttiD.gif" ALT="typeid D">
1287 
</CENTER>
1288 
</P>
1289 
 
1290 
<H4>Defining run-time type information structures</H4>
1291 
<P>
1292 
For built-in types, the run-time type information structure may be
1293 
referenced by the token:
1294 
<PRE>
1295 
	~cpp.typeid.basic : ( SIGNED_NAT ) -&gt; EXP pti
1296 
</PRE>
1297 
where the argument gives the encoding of the type as given in the
1298 
following table:
1299 
</P>
1300 
<CENTER>
1301 
<TABLE BORDER>
1302 
<TR><TH>Type</TH>   <TH>Encoding</TH>
1303 
<TH>Type</TH>   <TH>Encoding</TH>
1304 
<TR><TD ALIGN=CENTER>char</TD>  <TD ALIGN=CENTER>0</TD>
1305 
<TD ALIGN=CENTER>unsigned long</TD> <TD ALIGN=CENTER>11</TD>
1306 
<TR><TD ALIGN=CENTER>(error)</TD> <TD ALIGN=CENTER>1</TD>
1307 
<TD ALIGN=CENTER>float</TD>  <TD ALIGN=CENTER>12</TD>
1308 
<TR><TD ALIGN=CENTER>void</TD>  <TD ALIGN=CENTER>2</TD>
1309 
<TD ALIGN=CENTER>double</TD> <TD ALIGN=CENTER>13</TD>
1310 
<TR><TD ALIGN=CENTER>(bottom)</TD> <TD ALIGN=CENTER>3</TD>
1311 
<TD ALIGN=CENTER>long double</TD> <TD ALIGN=CENTER>14</TD>
1312 
<TR><TD ALIGN=CENTER>signed char</TD> <TD ALIGN=CENTER>4</TD>
1313 
<TD ALIGN=CENTER>wchar_t</TD> <TD ALIGN=CENTER>16</TD>
1314 
<TR><TD ALIGN=CENTER>signed short</TD> <TD ALIGN=CENTER>5</TD>
1315 
<TD ALIGN=CENTER>bool</TD>  <TD ALIGN=CENTER>17</TD>
1316 
<TR><TD ALIGN=CENTER>signed int</TD> <TD ALIGN=CENTER>6</TD>
1317 
<TD ALIGN=CENTER>(ptrdiff_t)</TD> <TD ALIGN=CENTER>18</TD>
1318 
<TR><TD ALIGN=CENTER>signed long</TD> <TD ALIGN=CENTER>7</TD>
1319 
<TD ALIGN=CENTER>(size_t)</TD> <TD ALIGN=CENTER>19</TD>
1320 
<TR><TD ALIGN=CENTER>unsigned char</TD> <TD ALIGN=CENTER>8</TD>
1321 
<TD ALIGN=CENTER>(...)</TD>  <TD ALIGN=CENTER>20</TD>
1322 
<TR><TD ALIGN=CENTER>unsigned short</TD><TD ALIGN=CENTER>9</TD>
1323 
<TD ALIGN=CENTER>signed long long</TD>
1324 
<TD ALIGN=CENTER>23</TD>
1325 
<TR><TD ALIGN=CENTER>unsigned int</TD> <TD ALIGN=CENTER>10</TD>
1326 
<TD ALIGN=CENTER>unsigned long long</TD>
1327 
<TD ALIGN=CENTER>27</TD>
1328 
</TABLE>
1329 
</CENTER>
1330 
<P>
1331 
Note that the encoding for the basic integral types is the same as
1332 
that
1333 
<A HREF="#arith">given above</A>.  The other types are assigned to
1334 
unused values.  Note that the encodings for <CODE>ptrdiff_t</CODE>
1335 
and
1336 
<CODE>size_t</CODE> are not used, instead that for their implementation
1337 
is used (using the standard tokens <CODE>ptrdiff_t</CODE> and
1338 
<CODE>size_t</CODE>).  The encodings for <CODE>bool</CODE> and
1339 
<CODE>wchar_t</CODE> are used because they are conceptually distinct
1340 
types even though they are implemented as one of the basic integral
1341 
types.  The type labelled <CODE>...</CODE> is the dummy used in the
1342 
representation of ellipsis functions.  The default implementation
1343 
uses an array of type information structures, <CODE>__TCPPLUS_typeid</CODE>,
1344 
to implement <CODE>~cpp.typeid.basic</CODE>.
1345 
</P>
1346 
<P>
1347 
The run-time type information structures for classes are defined in
1348 
the same place as their <A HREF="#vtable">virtual function tables</A>.
1349 
Other run-time type information structures are defined in whatever
1350 
modules require them.  In the former case the type information structure
1351 
will have an <A HREF="#other">external tag name</A>; in the latter
1352 
case it will be an internal tag.
1353 
</P>
1354 
 
1355 
<H4>Accessing run-time type information</H4>
1356 
<P>
1357 
The primary means of accessing the run-time type information for an
1358 
object is using the <CODE>typeid</CODE> construct.  In cases where
1359 
the operand type can be determined statically, the address of the
1360 
corresponding type information structure is returned.  In other cases
1361 
the token:
1362 
<PRE>
1363 
	~cpp.typeid.ref : ( EXP ppvt ) -&gt; EXP pti
1364 
</PRE>
1365 
is used, where the argument gives a reference to the <I>vptr</I> field
1366 
of the object being checked.  From this information it is trivial
1367 
to trace the corresponding type information.
1368 
</P>
1369 
<P>
1370 
Another means of querying the run-time type information for an object
1371 
is using the <CODE>dynamic_cast</CODE> construct.  When the result
1372 
cannot be determined statically, this is implemented using the token:
1373 
<PRE>
1374 
	~cpp.dynam.cast : ( EXP ppvt, EXP pti ) -&gt; EXP pv
1375 
</PRE>
1376 
where the first expression gives a reference to the <I>vptr</I> field
1377 
of the object being cast and the second gives the run-time type information
1378 
for the type being cast to.  In the default implementation this token
1379 
is implemented by the procedure <CODE>__TCPPLUS_dynamic_cast</CODE>.
1380 
The key point to note is that the virtual function table contains
1381 
the offset, <I>voff</I>, of the <I>vptr</I> field from the start of
1382 
the most complete object.  Thus it is possible to find the address
1383 
of the most complete object.  The run-time type information contains
1384 
enough information to determine whether this object has a sub-object
1385 
of the type being cast to, and if so, how to find the address of this
1386 
sub-object.  The result is returned as a <CODE>void *</CODE>, with
1387 
the null pointer indicating that the conversion is not possible.
1388 
</P>
1389 
 
1390 
<HR>
1391 
<H3><A NAME="init">2.6.15. Dynamic initialisation</A></H3>
1392 
<P>
1393 
The dynamic initialisation of variables with static storage duration
1394 
in C++ is implemented by means of the TDF <CODE>initial_value</CODE>
1395 
construct.  However in order for the producer to maintain control
1396 
over the order of initialisation, rather than each variable being
1397 
initialised separately using <CODE>initial_value</CODE>, a single
1398 
expression is created which initialises all the variables in a module,
1399 
and this initialiser expression is used to initialise a single dummy
1400 
variable using <CODE>initial_value</CODE>.  Note that, while this
1401 
enables the variables within a single module to be initialised in
1402 
the order in which they are defined, the order of initialisation between
1403 
different modules is unspecified.
1404 
</P>
1405 
<P>
1406 
The implementation needs to keep a list of those variables with static
1407 
storage duration which have been initialised so that it can call the
1408 
destructors for these objects at the end of the program. This is done
1409 
by declaring a variable of shape:
1410 
<PRE>
1411 
	~cpp.destr.type : () -&gt; SHAPE
1412 
</PRE>
1413 
for each such object with a non-trivial destructor.  Each element
1414 
of an array is considered a distinct object.  Immediately after the
1415 
variable has been initialised the token:
1416 
<PRE>
1417 
	~cpp.destr.global : ( EXP pd, EXP POINTER c, EXP PROC ) -&gt; EXP TOP
1418 
</PRE>
1419 
is called to add the variable to the list of objects to be destroyed.
1420 
The first argument is the address of the dummy variable just declared,
1421 
the second is the address of the object to be destroyed, and the third
1422 
is the destructor to be used.  In this way a list giving the objects
1423 
to be destroyed, and the order in which to destroy them, is built
1424 
up.  Note that partially constructed objects are destroyed within
1425 
their constructors (see <A HREF="#partial">above</A>) so that only
1426 
completely constructed objects need to be considered.
1427 
</P>
1428 
<P>
1429 
The implementation also needs to ensure that it calls the destructors
1430 
in this list at the end of the program, including calls of
1431 
<CODE>exit</CODE>.  This is done by calling the token:
1432 
<PRE>
1433 
	~cpp.destr.init : () -&gt; EXP TOP
1434 
</PRE>
1435 
at the start of each <CODE>initial_value</CODE> construct.  In the
1436 
default implementation this uses <CODE>atexit</CODE> to register a
1437 
function, <CODE>__TCPPLUS_term</CODE>, which calls the destructors.
1438 
To aid alternative implementations the token:
1439 
<PRE>
1440 
	~cpp.start : () -&gt; EXP TOP
1441 
</PRE>
1442 
is called at the start of the <CODE>main</CODE> function, however
1443 
this has no effect in the default implementation.
1444 
</P>
1445 
 
1446 
<HR>
1447 
<H3><A NAME="except">2.6.16. Exception handling</A></H3>
1448 
<P>
1449 
Conceptually, exception handling can be described in terms of the
1450 
following diagram:
1451 
<CENTER>
1452 
<IMG SRC="../images/try.gif" ALT="try stack">
1453 
</CENTER>
1454 
At any point in the execution of the program there is a stack of currently
1455 
active <CODE>try</CODE> blocks and currently active local variables.
1456 
A
1457 
<CODE>try</CODE> block is pushed onto the stack as it is entered and
1458 
popped from the stack when it is left (whether directly or via a jump).
1459 
A local variable with a non-trivial destructor is pushed onto the
1460 
stack just after its constructor has been called at the start of its
1461 
scope, and popped from the stack just before its destructor is called
1462 
at the end of its scope (including before jumps out of its scope).
1463 
Each element of an array is considered a separate object.  Each <CODE>try</CODE>
1464 
block has an associated list of handlers.  Each local variable has
1465 
an associated destructor.
1466 
</P>
1467 
<P>
1468 
Provided no exception is thrown this stack grows and shrinks in a
1469 
well-behaved manner as execution proceeds.  When an exception is thrown
1470 
an exception manager is invoked to find a matching exception handler.
1471 
The exception manager proceeds to execute a loop to unwind the stack
1472 
as follows.  If the stack is empty then the exception cannot be caught
1473 
and
1474 
<CODE>std::terminate</CODE> is called.  Otherwise the top element
1475 
is popped from the stack.  If this is a local variable then the associated
1476 
destructor is called for the variable.  If the top element is a
1477 
<CODE>try</CODE> block then the current exception is compared in turn
1478 
to each of the associated handlers.  If a match is found then execution
1479 
jumps to the handler body, otherwise the exception manager continues
1480 
to the next element of the stack.
1481 
</P>
1482 
<P>
1483 
Note that this description is purely conceptual.  There is no need
1484 
for exception handling to be implemented by a stack in this way (although
1485 
the default implementation uses a similar technique).  It does however
1486 
serve to illustrate the various stages which must exist in any implementation.
1487 
</P>
1488 
 
1489 
<H4>Try blocks</H4>
1490 
<P>
1491 
At the start of a <CODE>try</CODE> block a variable of shape:
1492 
<PRE>
1493 
	~cpp.try.type : () -&gt; SHAPE
1494 
</PRE>
1495 
is declared corresponding to the stack element for this block.  This
1496 
is then initialised using the token:
1497 
<PRE>
1498 
	~cpp.try.begin : ( EXP ptb, EXP POINTER fa, EXP POINTER ca ) -&gt; EXP TOP
1499 
</PRE>
1500 
</P>
1501 
where the first argument is a pointer to this variable, the second
1502 
argument is the TDF <CODE>current_env</CODE> construct, and the third
1503 
argument is the result of the TDF <CODE>make_local_lv</CODE> construct
1504 
on the label which is used to mark the first handler associated with
1505 
the block.  Note that the last two arguments enable a TDF
1506 
<CODE>long_jump</CODE> construct to be applied to transfer control
1507 
to the first handler.
1508 
<P>
1509 
When control exits from a <CODE>try</CODE> block, whether by reaching
1510 
the end of the block or jumping out of it, the block is removed from
1511 
the stack using the token:
1512 
<PRE>
1513 
	~cpp.try.end : ( EXP ptb ) -&gt; EXP TOP
1514 
</PRE>
1515 
where the argument is a pointer to the <CODE>try</CODE> block variable.
1516 
</P>
1517 
 
1518 
<H4>Local variables</H4>
1519 
<P>
1520 
The technique used to add a local variable with a non-trivial destructor
1521 
to the stack is similar to that used in the dynamic initialisation
1522 
of global variables.  A local variable of shape <CODE>~cpp.destr.type</CODE>
1523 
is declared at the start of the variable scope.  This is initialised
1524 
just after the constructor for the variable is called using the token:
1525 
<PRE>
1526 
	~cpp.destr.local : ( EXP pd, EXP POINTER c, EXP PROC ) -&gt; EXP TOP
1527 
</PRE>
1528 
where the first argument is a pointer to the variable being initialised,
1529 
the  second is a pointer to the local variable to be destroyed, and
1530 
the third is the destructor to be called.  At the end of the variable
1531 
scope, just before its destructor is called, the token:
1532 
<PRE>
1533 
	~cpp.destr.end : ( EXP pd ) -&gt; EXP TOP
1534 
</PRE>
1535 
where the argument is a pointer to destructor variable, is called
1536 
to remove the local variable destructor from the stack.  Note that
1537 
partially constructed objects are destroyed within their constructors
1538 
(see
1539 
<A HREF="#partial">above</A>) so that only completely constructed
1540 
objects need to be considered.
1541 
</P>
1542 
<P>
1543 
In cases where the local variable may be conditionally initialised
1544 
(for example a temporary variable in the second operand of a <CODE>||</CODE>
1545 
operation) the local variable of shape <CODE>~cpp.destr.type</CODE>
1546 
is initialised to the value given by the token:
1547 
<PRE>
1548 
	~cpp.destr.null : () -&gt; EXP d
1549 
</PRE>
1550 
(normally it is  left uninitialised).  Before the destructor for this
1551 
variable is called the value of the token:
1552 
<PRE>
1553 
	~cpp.destr.ptr : ( EXP pd ) -&gt; EXP POINTER c
1554 
</PRE>
1555 
is tested.  If <CODE>~cpp.destr.local</CODE> has been called for this
1556 
variable then this token returns a pointer to the variable, otherwise
1557 
it returns a null pointer.  The token <CODE>~cpp.destr.end</CODE>
1558 
and the destructor are only called if this token indicates that the
1559 
variable has been initialised.
1560 
</P>
1561 
 
1562 
<H4>Throwing an exception</H4>
1563 
<P>
1564 
When a <CODE>throw</CODE> expression with an argument is encountered
1565 
a number of steps performed.  Firstly, space is allocated to hold
1566 
the exception value using the token:
1567 
<PRE>
1568 
	~cpp.except.alloc : ( EXP VARIETY size_t ) -&gt; EXP pv
1569 
</PRE>
1570 
the argument of which gives the size of the value.  The space allocated
1571 
is returned as an expression of type <CODE>void *</CODE>.  Secondly,
1572 
the exception value is copied into the space allocated, using a copy
1573 
constructor if appropriate.  Finally the exception is raised using
1574 
the token:
1575 
<PRE>
1576 
	~cpp.except.throw : ( EXP pv, EXP pti, EXP PROC ) -&gt; EXP BOTTOM
1577 
</PRE>
1578 
The first argument gives the pointer to the exception value, returned
1579 
by
1580 
<CODE>~cpp.except.alloc</CODE>, the second argument gives a pointer
1581 
to the run-time type information for the exception type, and the third
1582 
argument gives the destructor to be called to destroy the exception
1583 
value (if any). This token sets the current exception to the given
1584 
values and invokes the exception manager as above.
1585 
</P>
1586 
<P>
1587 
A <CODE>throw</CODE> expression without an argument results in a call
1588 
to the token:
1589 
<PRE>
1590 
	~cpp.except.rethrow : () -&gt EXP BOTTOM
1591 
</PRE>
1592 
which re-invokes the exception manager with the current exception.
1593 
If there is no current exception then the implementation should call
1594 
<CODE>std::terminate</CODE>.
1595 
</P>
1596 
 
1597 
<H4>Handling an exception</H4>
1598 
<P>
1599 
The exception manager proceeds to find an exception in the manner
1600 
described above, unwinding the stack and calling destructors for local
1601 
variables.  When a <CODE>try</CODE> block is popped from the stack
1602 
a TDF <CODE>long_jump</CODE> is applied to transfer control to its
1603 
list of handlers.  For each handler in turn it is checked whether
1604 
the handler can catch the current exception.  For <CODE>...</CODE>
1605 
handlers this is always true; for other handlers it is checked using
1606 
the token:
1607 
<PRE>
1608 
	~cpp.except.catch : ( EXP pti ) -&gt; EXP VARIETY int
1609 
</PRE>
1610 
where the argument is a pointer to the run-time type information for
1611 
the handler type.  This token gives 1 if the exception is caught by
1612 
this handler, and 0 otherwise.  If the exception is not caught by
1613 
the handler then the next handler is checked, until there are no more
1614 
handlers associated with the <CODE>try</CODE> block.  In this case
1615 
control is passed back to the exception manager by re-throwing the
1616 
current exception using <CODE>~cpp.except.rethrow</CODE>.
1617 
</P>
1618 
<P>
1619 
If an exception is caught by a handler then a number of steps are
1620 
performed. Firstly, if appropriate, the handler variable is initialised
1621 
by copying the current exception value.  A pointer to the current
1622 
exception value can be obtained using the token:
1623 
<PRE>
1624 
	~cpp.except.value : () -&gt; EXP pv
1625 
</PRE>
1626 
Once this initialisation is complete the token:
1627 
<PRE>
1628 
	~cpp.except.caught : () -&gt; EXP TOP
1629 
</PRE>
1630 
is called to indicate that the exception has been caught.  The handler
1631 
body is then executed.  When control exits from the handler, whether
1632 
by reaching the end of the handler or by jumping out of it, the token:
1633 
<PRE>
1634 
	~cpp.except.end : () -&gt; EXP TOP
1635 
</PRE>
1636 
is called to indicate that the exception has been completed.  Note
1637 
that the implementation should call the destructor for the current
1638 
exception and free the space allocated by <CODE>~cpp.except.alloc</CODE>
1639 
at this point. Execution then continues with the statement following
1640 
the handler.
1641 
</P>
1642 
<P>
1643 
To conclude, the TDF generated for a <CODE>try</CODE> block and its
1644 
associated list of handlers has the form:
1645 
<PRE>
1646 
	variable (
1647 
	    long_jump_access,
1648 
	    stack_tag,
1649 
	    make_value ( ~cpp.try.type ),
1650 
	    conditional (
1651 
		handler_label,
1652 
		sequence (
1653 
		    ~cpp.try.begin (
1654 
			obtain_tag ( stack_tag ),
1655 
			current_env,
1656 
			make_local_lv ( handler_label ) ),
1657 
			<I>try-block-body</I>,
1658 
			~cpp.try.end ),
1659 
		    conditional (
1660 
			catch_label_1,
1661 
			sequence (
1662 
			    integer_test (
1663 
				not_equal,
1664 
				catch_label_1,
1665 
				~cpp.except.catch (
1666 
				    <I>handler-1-typeid</I> ) )
1667 
			    variable (
1668 
				handler_tag_1,
1669 
				<I>handler-1-init</I> (
1670 
				    ~cpp.except.value ),
1671 
				sequence (
1672 
				    ~cpp.except.caught,
1673 
				    <I>handler-1-body</I> ) )
1674 
			    ~cpp.except.end )
1675 
			conditional (
1676 
			    catch_label_2,
1677 
			    <I>further-handlers</I>,
1678 
			    ~cpp.except.rethrow ) ) ) )
1679 
</PRE>
1680 
</P>
1681 
<P>
1682 
Note that for a local variable to maintain its previous value when
1683 
an  exception is caught in this way it is necessary to declare it
1684 
using the TDF <CODE>long_jump_access</CODE> construct.  Any local
1685 
variable which contains a <CODE>try</CODE> block in its scope is declared
1686 
in this way.
1687 
</P>
1688 
<P>
1689 
To aid implementations in the writing of exception managers the following
1690 
standard tokens are provided:
1691 
<PRE>
1692 
	~cpp.ptr.code : () -&gt; SHAPE POINTER ca
1693 
	~cpp.ptr.frame : () -&gt; SHAPE POINTER fa
1694 
	~cpp.except.jump : ( EXP POINTER fa, EXP POINTER ca ) -&gt; EXP BOTTOM
1695 
</PRE>
1696 
These give the shape of the TDF <CODE>make_local_lv</CODE> construct,
1697 
the shape of the TDF <CODE>current_env</CODE> construct, and direct
1698 
access to the TDF <CODE>long_jump</CODE> access.  The exception manager
1699 
in the default implementation is a function called <CODE>__TCPPLUS_throw</CODE>.
1700 
</P>
1701 
 
1702 
<H4>Exception specifications</H4>
1703 
<P>
1704 
If a function is declared with an exception specification then extra
1705 
code needs to be generated in the function definition to catch any
1706 
unexpected exceptions thrown by the function and to call <CODE>std::unexpected
1707 
</CODE>. Since this is a potentially high overhead for small functions,
1708 
this extra code is not generated if it can be proved that such unexpected
1709 
exceptions can never be thrown (the analysis is essentially the same
1710 
as that in the
1711 
<A HREF="pragma.html#exception">exception analysis</A> check).
1712 
</P>
1713 
<P>
1714 
The implementation of exception specification is to enclose the entire
1715 
function definition in a <CODE>try</CODE> block.  The handler for
1716 
this block uses <CODE>~cpp.except.catch</CODE> to check whether the
1717 
current exception can be caught by any of the types listed in the
1718 
exception specification.  If so the current exception is re-thrown.
1719 
If none of these types catch the current exception then the token:
1720 
<PRE>
1721 
	~cpp.except.bad : ( SIGNED_NAT ) -&gt; EXP TOP
1722 
</PRE>
1723 
is called.  The argument is 1 if the exception specification includes
1724 
the special type <CODE>std::bad_exception</CODE>, and 0 otherwise.
1725 
The implementation should call <CODE>std::unexpected</CODE>, but how
1726 
any exceptions thrown during this call are to be handled depends on
1727 
the value of the argument.
1728 
</P>
1729 
 
1730 
<HR>
1731 
<H3><A NAME="mangle">2.6.17. Mangled identifier names</A></H3>
1732 
<P>
1733 
In a similar fashion to other C++ compilers, the C++ producer needs
1734 
a method of mapping C++ identifiers to a form suitable for further
1735 
processing, namely TDF tag names.  This mangled name contains an encoding
1736 
of the identifier name, its parent namespace or class and its type.
1737 
Identifiers with C linkage are not mangled.  The producer contains
1738 
a built-in <A HREF="man.html#unmangle">name unmangler</A>
1739 
which performs the reverse operation of transforming the mangled form
1740 
of an identifier name back to the underlying identifier.  This can
1741 
be useful when analysing system linker errors.
1742 
</P>
1743 
<P>
1744 
Note that the type of an identifier forms part of its mangled name
1745 
not only for functions, but also for variables.  Many other compilers
1746 
do not mangle variable names, however the ISO C++ rules on namespaces
1747 
and variables with C linkage make it necessary (this can be suppressed
1748 
using the <CODE>-j-n</CODE> command-line option).  Declaring the language
1749 
linkage of a variable inconsistently can therefore lead to linking
1750 
errors with the C++ producer which are not detected by other compilers.
1751 
A common example is:
1752 
<PRE>
1753 
	extern int errno ;
1754 
</PRE>
1755 
which, leaving aside whether <CODE>errno</CODE> is actually an external
1756 
variable, should be:
1757 
<PRE>
1758 
	extern &quot;C&quot; int errno ;
1759 
</PRE>
1760 
</P>
1761 
<P>
1762 
As described above, the mangled form of an identifier has three components;
1763 
the identifier name, the identifier namespace and the identifier type.
1764 
Two underscores (<CODE>__</CODE>) are used to separate the name component
1765 
from the namespace and type components.  The mangling scheme used
1766 
is based on that described in the ARM.  The description below is not
1767 
complete; the mangling and unmangling routines themselves should be
1768 
consulted for a complete description.
1769 
</P>
1770 
 
1771 
<H4>Mangling identifier names</H4>
1772 
<P>
1773 
Simple identifier names are mapped to themselves.  Unicode characters
1774 
of the forms <CODE>\u</CODE><I>xxxx</I> and <CODE>\U</CODE><I>xxxxxxxx</I>
1775 
are mapped to <CODE>__k</CODE><I>xxxx</I> and <CODE>__K</CODE><I>xxxxxxxx</I>
1776 
respectively, where the hex digits are output in their canonical lower-case
1777 
form.  Constructors are mapped to <CODE>__ct</CODE> and destructors
1778 
to <CODE>__dt</CODE>.  Conversions functions are mapped to
1779 
<CODE>__op</CODE><I>type</I> where <I>type</I> is the mangled form
1780 
of the conversion type.  Overloaded operator functions,
1781 
<CODE>operator@</CODE>, are mapped as follows:
1782 
</P>
1783 
<CENTER>
1784 
<TABLE BORDER>
1785 
<TR><TH>Operator</TH>   <TH>Mapping</TH>
1786 
<TH>Operator</TH>   <TH>Mapping</TH>
1787 
<TH>Operator</TH>   <TH>Mapping</TH>
1788 
<TR><TD ALIGN=CENTER>&amp;</TD>  <TD ALIGN=CENTER>__ad</TD>
1789 
<TD ALIGN=CENTER>&amp;=</TD> <TD ALIGN=CENTER>__aad</TD>
1790 
<TD ALIGN=CENTER>[]</TD>  <TD ALIGN=CENTER>__vc</TD>
1791 
<TR><TD ALIGN=CENTER>-&gt;</TD>  <TD ALIGN=CENTER>__rf</TD>
1792 
<TD ALIGN=CENTER>-&gt;*</TD> <TD ALIGN=CENTER>__rm</TD>
1793 
<TD ALIGN=CENTER>=</TD>  <TD ALIGN=CENTER>__as</TD>
1794 
<TR><TD ALIGN=CENTER>,</TD>  <TD ALIGN=CENTER>__cm</TD>
1795 
<TD ALIGN=CENTER>~</TD>  <TD ALIGN=CENTER>__co</TD>
1796 
<TD ALIGN=CENTER>/</TD>  <TD ALIGN=CENTER>__dv</TD>
1797 
<TR><TD ALIGN=CENTER>/=</TD>  <TD ALIGN=CENTER>__adv</TD>
1798 
<TD ALIGN=CENTER>==</TD>  <TD ALIGN=CENTER>__eq</TD>
1799 
<TD ALIGN=CENTER>()</TD>  <TD ALIGN=CENTER>__cl</TD>
1800 
<TR><TD ALIGN=CENTER>&gt;</TD>  <TD ALIGN=CENTER>__gt</TD>
1801 
<TD ALIGN=CENTER>&gt;=</TD>  <TD ALIGN=CENTER>__ge</TD>
1802 
<TD ALIGN=CENTER>&lt;</TD>  <TD ALIGN=CENTER>__lt</TD>
1803 
<TR><TD ALIGN=CENTER>&lt;=</TD>  <TD ALIGN=CENTER>__le</TD>
1804 
<TD ALIGN=CENTER>&amp;&amp;</TD> <TD ALIGN=CENTER>__aa</TD>
1805 
<TD ALIGN=CENTER>||</TD>  <TD ALIGN=CENTER>__oo</TD>
1806 
<TR><TD ALIGN=CENTER>&lt;&lt;</TD> <TD ALIGN=CENTER>__ls</TD>
1807 
<TD ALIGN=CENTER>&lt;&lt;=</TD> <TD ALIGN=CENTER>__als</TD>
1808 
<TD ALIGN=CENTER>-</TD>  <TD ALIGN=CENTER>__mi</TD>
1809 
<TR><TD ALIGN=CENTER>-=</TD>  <TD ALIGN=CENTER>__ami</TD>
1810 
<TD ALIGN=CENTER>--</TD>  <TD ALIGN=CENTER>__mm</TD>
1811 
<TD ALIGN=CENTER>!</TD>  <TD ALIGN=CENTER>__nt</TD>
1812 
<TR><TD ALIGN=CENTER>!=</TD>  <TD ALIGN=CENTER>__ne</TD>
1813 
<TD ALIGN=CENTER>|</TD>  <TD ALIGN=CENTER>__or</TD>
1814 
<TD ALIGN=CENTER>|=</TD>  <TD ALIGN=CENTER>__aor</TD>
1815 
<TR><TD ALIGN=CENTER>+</TD>  <TD ALIGN=CENTER>__pl</TD>
1816 
<TD ALIGN=CENTER>+=</TD>  <TD ALIGN=CENTER>__apl</TD>
1817 
<TD ALIGN=CENTER>++</TD>  <TD ALIGN=CENTER>__pp</TD>
1818 
<TR><TD ALIGN=CENTER>%</TD>  <TD ALIGN=CENTER>__md</TD>
1819 
<TD ALIGN=CENTER>%=</TD>  <TD ALIGN=CENTER>__amd</TD>
1820 
<TD ALIGN=CENTER>&gt;&gt;</TD> <TD ALIGN=CENTER>__rs</TD>
1821 
<TR><TD ALIGN=CENTER>&gt;&gt;=</TD> <TD ALIGN=CENTER>__ars</TD>
1822 
<TD ALIGN=CENTER>*</TD>  <TD ALIGN=CENTER>__ml</TD>
1823 
<TD ALIGN=CENTER>*=</TD>  <TD ALIGN=CENTER>__aml</TD>
1824 
<TR><TD ALIGN=CENTER>^</TD>  <TD ALIGN=CENTER>__er</TD>
1825 
<TD ALIGN=CENTER>^=</TD>  <TD ALIGN=CENTER>__aer</TD>
1826 
<TD ALIGN=CENTER>delete</TD> <TD ALIGN=CENTER>__dl</TD>
1827 
<TR><TD ALIGN=CENTER>delete []</TD> <TD ALIGN=CENTER>__vd</TD>
1828 
<TD ALIGN=CENTER>new</TD>  <TD ALIGN=CENTER>__nw</TD>
1829 
<TD ALIGN=CENTER>new []</TD> <TD ALIGN=CENTER>__vn</TD>
1830 
<TR><TD ALIGN=CENTER>?:</TD>  <TD ALIGN=CENTER>__cn</TD>
1831 
<TD ALIGN=CENTER>:</TD>  <TD ALIGN=CENTER>__cs</TD>
1832 
<TD ALIGN=CENTER>::</TD>  <TD ALIGN=CENTER>__cc</TD>
1833 
<TR><TD ALIGN=CENTER>.</TD>  <TD ALIGN=CENTER>__df</TD>
1834 
<TD ALIGN=CENTER>.*</TD>  <TD ALIGN=CENTER>__dm</TD>
1835 
<TD ALIGN=CENTER>abs</TD>  <TD ALIGN=CENTER>__ab</TD>
1836 
<TR><TD ALIGN=CENTER>max</TD>  <TD ALIGN=CENTER>__mx</TD>
1837 
<TD ALIGN=CENTER>min</TD>  <TD ALIGN=CENTER>__mn</TD>
1838 
<TD ALIGN=CENTER>sizeof</TD> <TD ALIGN=CENTER>__sz</TD>
1839 
<TR><TD ALIGN=CENTER>typeid</TD> <TD ALIGN=CENTER>__td</TD>
1840 
<TD ALIGN=CENTER>vtable</TD> <TD ALIGN=CENTER>__tb</TD>
1841 
<TD ALIGN=CENTER>-</TD>  <TD ALIGN=CENTER>-</TD>
1842 
</TABLE>
1843 
</CENTER>
1844 
<P>
1845 
Note that this table contains a number of operators which are not
1846 
part of C++ or cannot be overloaded in C++.  These are used in the
1847 
representation of target dependent integer constants.
1848 
</P>
1849 
 
1850 
<H4>Mangling namespace names</H4>
1851 
<P>
1852 
The global namespace is mapped to an empty string.  Simple namespace
1853 
and class names are mapped as above, but are preceded by a series
1854 
of decimal digits giving the length of the mangled name.  Nested namespaces
1855 
and classes are represented by a sequence of such namespace names,
1856 
preceded by the number of elements in the sequence.  This takes the
1857 
form <CODE>Q</CODE><I>digit</I> if there are less than 10 elements,
1858 
or
1859 
<CODE>Q_</CODE><I>digits</I><CODE>_</CODE> if there are more than
1860 
10. Note that members of anonymous classes or namespaces are local
1861 
to their translation unit, and so do not have external tag names.
1862 
</P>
1863 
 
1864 
<H4>Mangling types</H4>
1865 
<P>
1866 
The mangling of types is essentially similar to that used in the
1867 
<A HREF="dump.html">symbol table dump</A> format.  The type used in
1868 
the mangled name for an identifier ignores the return type for a function
1869 
and ignores the most significant bound for an array.
1870 
</P>
1871 
<P>
1872 
The built-in types are mapped in precisely the same way as in the
1873 
<A HREF="dump.html#built-in">symbol table dump</A>.  Class and enumeration
1874 
types are mapped to their type names mangled in the same way as the
1875 
namespace names above.  The exception to this is that in a class member,
1876 
the parent class is mapped to <CODE>X</CODE>.
1877 
</P>
1878 
<P>
1879 
The composite types are again mapped in a similar fashion to that
1880 
in the <A HREF="dump.html#composite">dump file</A>.  For example,
1881 
<CODE>PCc</CODE> represents <CODE>const char *</CODE>.  The only difficult
1882 
case concerns function parameter types where the ARM
1883 
<CODE>T</CODE> and <CODE>N</CODE> encodings are used for duplicate
1884 
parameter types.  The function return type is included in the mangled
1885 
form except for function identifier types.  In the cases where the
1886 
identifier is known always to represent a function (constructors,
1887 
destructors etc.) the initial <CODE>F</CODE>
1888 
indicating a function type is also omitted.
1889 
</P>
1890 
<P>
1891 
The types of template functions and classes are represented by the
1892 
underlying template and the template arguments giving rise to the
1893 
instance.  Template classes are preceded by <CODE>t</CODE>; template
1894 
functions are preceded by <CODE>G</CODE> rather than <CODE>F</CODE>.
1895 
Type arguments are represented by <CODE>Z</CODE> followed by the type
1896 
value; non-type arguments are represented by the argument type followed
1897 
by the argument value.  In the underlying type the template parameters
1898 
are represented by <CODE>m0</CODE>, <CODE>m1</CODE> etc. An alternative
1899 
scheme, in which the mangled form of a template function includes
1900 
the type of that instance, rather than the underlying template, can
1901 
be enabled using the <CODE>-j-f</CODE>
1902 
command-line option.
1903 
</P>
1904 
 
1905 
<H4><A NAME="other">Other mangled names</A></H4>
1906 
<P>
1907 
The <A HREF="#vtable">virtual function table</A> for a class, when
1908 
this is a variable with external linkage, is named <CODE>__vt__</CODE><I>type
1909 
</I>, where <I>type</I> is the mangled form of the class name.  The
1910 
virtual function table for a base class is named <CODE>__vt__</CODE><I>base</I>
1911 
where <I>base</I> is a sequence of mangled class names specifying
1912 
the base class.  The <A HREF="#rtti">run-time type information structure</A>
1913 
for a type, when this is a variable with external linkage, is named
1914 
<CODE>__ti__</CODE><I>type</I>, where <I>type</I> is the mangled form
1915 
of the type name.
1916 
</P>
1917 
 
1918 
<H4>Mangled name examples</H4>
1919 
<P>
1920 
The following gives some examples of the name mangling scheme:
1921 
<PRE>
1922 
	class A {
1923 
	    static int a ;			// a__1Ai
1924 
	public :
1925 
	    A () ;				// __ct__1A
1926 
	    A ( int ) ;				// __ct__1Ai
1927 
	    A ( const A &amp; ) ;			// __ct__1ARCX
1928 
	    virtual ~A () ;			// __dt__1A
1929 
	    operator bool () ;			// __opb__1A
1930 
	    bool operator! () ;			// __nt__1A
1931 
	} ;
1932 
 
1933 
	// virtual function table	__vt__1A
1934 
	// run-time type information	__ti__1A
1935 
 
1936 
	int f ( A *, int, A * ) ;		// f__FP1AiT1
1937 
	int b = 2 ;				// b__i
1938 
	int c [3] ;				// c__A_i
1939 
 
1940 
	namespace N {
1941 
	    int *p = 0 ;			// p__1NPi
1942 
	}
1943 
</PRE>
1944 
</P>
1945 
 
1946 
<HR>
1947 
<P><I>Part of the <A HREF="../index.html">TenDRA Web</A>.<BR>Crown
1948 
Copyright &copy; 1998.</I></P>
1949 
</BODY>
1950 
</HTML>