[RFA] backtrace in mixed language applications (take 2)

[Joel Brobecker <brobecker at adacore dot com>]

Hello,

This is a problem impacting multi-language applications that Jerome
reported a while ago:

On Fri, Aug 10, 2007 at 12:03:13PM +0200, Jerome Guitton wrote:
> when doing a backtrace in a mixed-language application, GDB should use the
> appriorate language for printing the frames (in particular for printing
> arguments). Not the case for now, it prints every frame with the
> language of the first one.
> 
> This is quite visible in a mixed C/Ada application; for example, in
> Ada, pointer to unconstrained strings can be represented with "fat pointers",
> which are records containing a pointer to the bound information (P_BOUNDS)
> and a pointer to the array of characters (P_ARRAY). Imagine that you have
> an ada procedure lang_switch.ada_procedure, which takes an Ada string in
> parameter, calling some C code. If you get a backtrace from the C code,
> you'll get:
> 
> [...]
> #0  c_procedure (msg=0xbfffc170 "msg") at foo.c:4
> #1  0x08049b91 in lang_switch.ada_procedure (msg={P_ARRAY = 0x805524c,
>     P_BOUNDS = 0x8055250}) at lang_switch.adb:14
> [...]
> 
> the msg is bogus in frame #1 is bogus. you should have got:
> 
> [...]
> #0  c_procedure (msg=0xbfffaa00 "msg") at foo.c:4
> #1  0x08049b91 in lang_switch.ada_procedure (msg=0x805524c)
>     at lang_switch.adb:14
> [...]

Jerome posted a patch which I reviewed, and we decided that the patch
needed to be reworked. Since then, Jerome has been quite busy and
asked me to do it for him.

I looked into the problem some more, and found that we could quite
easily avoid switching a global variable, and instead find the
appropriate language and explicitly pass that language to the
val_print routine.  This implies the addition of the language
to a couple of routines (common_val_print and val_print), and
a lot of mechanical edits all over, but it's a good step in the
right direction anyway.

As a side note, I am still thinking about updating the *parsing*
routines to take a new struct parse_context parameter. In this
structure, we will have the input_radix in addition to the language.
I considered using that same structure for the *printing* but
as a matter for fact, only the language is relevant in the printing
case. So I only added the language.

The meat of the change is in valprint.c (update of the val_print
and common_val_print function profiles, as well as removing the
use of the current_language), in stack.c (where we get the language
from the symbol to be printed - we could get it from the frame but
I think it's the same), and mi-stack.c. The rest are mechanical
edits where I added the current_language. I didn't want to look
at each of the call sites individually and try to guess whether
we can pass something better. I'd much rather leave that for later,
as a series of separate patches.

2008-01-09  Joel Brobecker  <brobecker@adacore.com>

        * valprint.c (val_print): Add new language parameter and use it
        instead of using the current_language. Update calls to val_print
        throughout.
        (common_val_print): Add new langauge parameter and pass it to
        val_print.
        * value.h (struct language_defn): Add opaque declaration.
        (val_print, common_val_print): Update declarations.
        * stack.c (print_frame_args): Update call to common_val_print
        using the appropriate language.
        * mi/mi-cmd-stack.c (list_args_or_locals): Likewise.
        * c-valprint, f-valprint.c, m2-valprint.c, mt-tdep.c, infcmd.c,
        mi/mi-main.c, jv-valprint.c, ada-valprint.c, varobj.c, p-valprint.c,
        scm-valprint.c, cp-valprint.c, sh64-tdep.c, printcmd.c: 
        #include "language.h" if necessary. 
        Update calls to val_print and common_val_print.
        * Makefile.in (mt-tdep.o, sh64-tdep.o, mi-cmds.o, mi-main.o):
        Update dependencies.

The testcase is still pretty much the same. I just fixed a couple
of errors that show up when testing out of tree.

2008-01-09  Jerome Guitton  <guitton@adacore.com>

        * gdb.ada/lang_switch: New test program.
        * gdb.ada/lang_switch.exp: New testcase.

Tested on x86-linux. OK to apply?

Thanks,
-- 
Joel

Attachment: lang_switch.diff
Description: Text document

Attachment: lang_switch-tc.diff
Description: Text document