[RFA] Fix pascal behavior for class fields

[RFA] Fix pascal behavior for class fields

[Pierre Muller]

  This is the problem that triggered my previous RFA,
as I discovered it while debugging this problem.

  In the pascal parser, there is special code
that will try to emulate case-insensivity despite the
fact that pascal is not registered as a 'case-insensitive' language.

  This code does not work as expected for fields of a class,
for which exact casing is currently required.

See:
https://sourceware.org/bugzilla/show_bug.cgi?id=17815
for source code:

# Compile attached source code, using Free Pascal compiler

fpc -gs -Mobjfpc test-class-pascal.pas

# Debug it
gdb ./test-class-pascal
.......
(gdb) b TA__CHECK
Breakpoint 1 at 0x40154c: file test-class-pascal.pas, line 23.
(gdb) r
Starting program: E:\pas\test\test-class-pascal.exe
[New Thread 6700.0x1ae0]

Breakpoint 1, TA__CHECK (B=0x1572ee0, this=<error reading variable>)
    at test-class-pascal.pas:23
23        check:=(x < b.x);
(gdb) p this
$1 = (TA) 0x1572ed0
(gdb) p this^
warning: can't find linker symbol for virtual table for `TA' value
$2 = {<TOBJECT> = {_vptr$ = {0x408014, 0x43}}, X = 67, Y = 33}
(gdb) p X
warning: can't find linker symbol for virtual table for `TA' value
$3 = 67
(gdb) p B.X
warning: can't find linker symbol for virtual table for `TA' value
$4 = -1
(gdb) p b.x
warning: can't find linker symbol for virtual table for `TA' value
$5 = -1
(gdb) p x
Type TA has no component named x.
(gdb)


  The patch below fixes this odd behavior,
if there is no approval nor comments about it,
I will commit it in a few days (before 7.9 branching)
as pascal language maintainer.

  But, of course, don't hesitate to give feedback!
 
Pierre Muller

2015-01-07  Pierre Muller  <muller@sourceware.org>

       PR pascal/17815
       * p-exp.y (yylex): Remember the case pattern that allowed finding
       a field of this.


diff --git a/gdb/p-exp.y b/gdb/p-exp.y
index a1c78bf..3a4905a 100644
--- a/gdb/p-exp.y
+++ b/gdb/p-exp.y
@@ -1598,7 +1598,7 @@ yylex (void)
                              VAR_DOMAIN, &is_a_field_of_this);
       }

-    if (is_a_field)
+    if (is_a_field || (is_a_field_of_this.type != NULL))
       {
        tempbuf = (char *) realloc (tempbuf, namelen + 1);
        strncpy (tempbuf, tmp, namelen);
@@ -1606,7 +1606,11 @@ yylex (void)
        yylval.sval.ptr = tempbuf;
        yylval.sval.length = namelen;
        free (uptokstart);
-       return FIELDNAME;
+        yylval.ssym.is_a_field_of_this = is_a_field_of_this.type != NULL;
+       if (is_a_field)
+         return FIELDNAME;
+       else
+         return NAME;
       }
     /* Call lookup_symtab, not lookup_partial_symtab, in case there are
        no psymtabs (coff, xcoff, or some future change to blow away the
@@ -1739,7 +1743,6 @@ yylex (void)
     free(uptokstart);
     /* Any other kind of symbol.  */
     yylval.ssym.sym = sym;
-    yylval.ssym.is_a_field_of_this = is_a_field_of_this.type != NULL;
     return NAME;
   }
 }

Re: [RFA] Fix pascal behavior for class fields

[Pedro Alves]

On 01/08/2015 01:00 PM, Pierre Muller wrote:
>> De : gdb-patches-owner@sourceware.org [mailto:gdb-patches-
>> owner@sourceware.org] De la part de Pedro Alves
>> Envoyé : jeudi 8 janvier 2015 11:48
>> À : Pierre Muller; 'gdb-patches'
>> Objet : Re: [RFA] Fix pascal behavior for class fields
>>> (gdb) p b.x
>>> warning: can't find linker symbol for virtual table for `TA' value
>>> $5 = -1
>>> (gdb) p x
>>> Type TA has no component named x.
>>> (gdb)
>>
>> How about adding this to the test suite?
> 
> 
>   The whole testsuite/gpc.pascal is almost empty,
> I never invested time to develop it :(
> 
>   At the time I started it, GPC (the GNU pascal compiler)
> was still active, but development apparently
> stopped since quite some time.
> 
>   I am unable to install GPC, which means that I cannot test it.
> Would a testsuite that supports only Free Pascal be acceptable?

Do you actually mean, whether it's ok for a new
test (not test suite) to go in untested on GPC?  It certainly is.
Better test on FPC than nowhere.  :-)

AFAICS, the tests themselves don't really care which compiler
is in use other than for marking xfails; you just call gdb_compile_pascal,
and that works with either.  That's my impression from quickly
skimming testsuite/lib/pascal.exp.

Thanks,
Pedro Alves

RE: [RFA] Fix pascal behavior for class fields

[Pierre Muller]

> De : gdb-patches-owner@sourceware.org [mailto:gdb-patches-
> owner@sourceware.org] De la part de Pedro Alves
> Envoyé : jeudi 8 janvier 2015 11:48
> À : Pierre Muller; 'gdb-patches'
> Objet : Re: [RFA] Fix pascal behavior for class fields
> > (gdb) p b.x
> > warning: can't find linker symbol for virtual table for `TA' value
> > $5 = -1
> > (gdb) p x
> > Type TA has no component named x.
> > (gdb)
> 
> How about adding this to the test suite?


  The whole testsuite/gpc.pascal is almost empty,
I never invested time to develop it :(

  At the time I started it, GPC (the GNU pascal compiler)
was still active, but development apparently
stopped since quite some time.

  I am unable to install GPC, which means that I cannot test it.
Would a testsuite that supports only Free Pascal be acceptable?

Pierre Muller
as GDB pascal language maintainer

Re: [RFA] Fix pascal behavior for class fields

[Pedro Alves]

On 01/08/2015 08:53 AM, Pierre Muller wrote:
>   This is the problem that triggered my previous RFA,
> as I discovered it while debugging this problem.
> 
>   In the pascal parser, there is special code
> that will try to emulate case-insensivity despite the
> fact that pascal is not registered as a 'case-insensitive' language.
> 
>   This code does not work as expected for fields of a class,
> for which exact casing is currently required.
> 
> See:
> https://sourceware.org/bugzilla/show_bug.cgi?id=17815
> for source code:
> 
> # Compile attached source code, using Free Pascal compiler
> 
> fpc -gs -Mobjfpc test-class-pascal.pas
> 
> # Debug it
> gdb ./test-class-pascal
> .......
> (gdb) b TA__CHECK
> Breakpoint 1 at 0x40154c: file test-class-pascal.pas, line 23.
> (gdb) r
> Starting program: E:\pas\test\test-class-pascal.exe
> [New Thread 6700.0x1ae0]
> 
> Breakpoint 1, TA__CHECK (B=0x1572ee0, this=<error reading variable>)
>     at test-class-pascal.pas:23
> 23        check:=(x < b.x);
> (gdb) p this
> $1 = (TA) 0x1572ed0
> (gdb) p this^
> warning: can't find linker symbol for virtual table for `TA' value
> $2 = {<TOBJECT> = {_vptr$ = {0x408014, 0x43}}, X = 67, Y = 33}
> (gdb) p X
> warning: can't find linker symbol for virtual table for `TA' value
> $3 = 67
> (gdb) p B.X
> warning: can't find linker symbol for virtual table for `TA' value
> $4 = -1
> (gdb) p b.x
> warning: can't find linker symbol for virtual table for `TA' value
> $5 = -1
> (gdb) p x
> Type TA has no component named x.
> (gdb)

How about adding this to the test suite?

Thanks,
Pedro Alves