This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: [patch] [gdb/testsuite] include a use of the definition of a type to cause clang to emit debug info
- From: Eric Christopher <echristo at gmail dot com>
- To: Joel Brobecker <brobecker at adacore dot com>
- Cc: David Blaikie <dblaikie at gmail dot com>, gdb-patches <gdb-patches at sourceware dot org>
- Date: Mon, 14 Apr 2014 08:53:37 -0700
- Subject: Re: [patch] [gdb/testsuite] include a use of the definition of a type to cause clang to emit debug info
- Authentication-results: sourceware.org; auth=none
- References: <CAENS6EvK8+AUQrj_0q2vAz0JH-5Pt0Z=cRWtW--4pyyUTP3WfQ at mail dot gmail dot com> <20140414131050 dot GR4250 at adacore dot com>
On Mon, Apr 14, 2014 at 6:10 AM, Joel Brobecker <brobecker@adacore.com> wrote:
> On Sat, Apr 12, 2014 at 11:43:40PM -0700, David Blaikie wrote:
>> Clang has an optimization that causes a the debug info to only include
>> the declaration of a type if the type is referenced but never used in
>> a context that requires a definition (eg: pointers are handed around
>> but never deferenced).
>>
>> This patch introduces a variable to one test file to cause clang to
>> emit the full definition of the type as well as fixing up a related
>> typo in the test message of the associated expect file.
>>
>> Like the difference between GCC and Clang in the emission of unused
>> static entities, I think this case is also a matter of degrees - both
>> GCC and Clang implement other similar optimizations* to the one
>> outlined here and the GDB test suite has managed to pass without
>> disabling those optimizations in GCC and I hope it's suitable to do
>> the same for Clang.
>>
>> Though admittedly I don't have much of the context of the history of
>> the testsuite, its priorities/preferences when it comes to
>> distinguishing testing compiler behavior versus debugger behavior,
>> etc.
>>
>> * the one I know of involves dynamic types: both GCC and Clang only
>> emit the debug info definition of such a type in any translation unit
>> that emits the key function. This means in many contexts where a full
>> definition is provided in the source only a declaration is provided in
>> the debug info.
>
>> commit 1128f6fb45483d45668d09e0696f4a590334e0c4
>> Author: David Blaikie <dblaikie@gmail.com>
>> Date: Sat Apr 12 23:27:19 2014 -0700
>>
>> Cause clang to emit the definition of a type used only by pointer
>>
>> gdb/testsuite/
>> * gdb.stabs/gdb11479.c: introduce a variable to cause clang to
>> emit the full definition of type required by the test
>> * gdb.stabs/gdb11479.exp: correct a typo in a test message
>
> The gdb11479.exp is obvious and can be pushed. The change to gdb11479.c,
> on the other hand, changes the test, and I don't think we want that,
> because I disagree with the outcome of Clang's optimization here.
> If Clang doesn't want to fix the problem, best to just xfail the test
> with Clang, and write a new one that provides the type definition as
> Clang wants it.
>
Could you describe your objection here? In particular, as it relates
to Dave's analysis of a similar gcc optimization. (Also a few people
have been interested in implementing this same behavior in gcc).
-eric