Bug 97394 - Incorrect analyzer output for setjmp
Summary: Incorrect analyzer output for setjmp
Status: RESOLVED FIXED
Alias: None
Product: gcc
Classification: Unclassified
Component: analyzer (show other bugs)
Version: 10.2.0
: P3 normal
Target Milestone: ---
Assignee: David Malcolm
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2020-10-13 02:14 UTC by Michael Morrell
Modified: 2020-10-14 20:55 UTC (History)
0 users

See Also:
Host:
Target:
Build:
Known to work:
Known to fail:
Last reconfirmed:

Attachments
Header file (316 bytes, text/x-csrc)
2020-10-13 02:14 UTC, Michael Morrell 		Details
Source file (131 bytes, text/x-csrc)
2020-10-13 02:15 UTC, Michael Morrell 		Details
Expected output (178 bytes, text/plain)
2020-10-13 02:15 UTC, Michael Morrell 		Details
Incorrect output (178 bytes, text/plain)
2020-10-13 02:16 UTC, Michael Morrell 		Details
[PATCH] analyzer: don't use <setjmp.h> in tests [PR97394] (1.06 KB, patch)
2020-10-13 20:38 UTC, David Malcolm 		Details | Diff
View All Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this bug.
Description Michael Morrell 2020-10-13 02:14:42 UTC 
Created attachment 49356 [details]
Header file

While testing gcc 10.2 with MUSL for libc, I noticed some of analyzer setjmp tests in the gcc validation suite were failing.

I managed to get a somewhat reduced testcase and determined that the problem was due to this #define in setjmp.h for MUSL:

#define setjmp setjmp

In glibc, the corresponding #define is:

#define setjmp(env) _setjmp(env)

and the testcases pass when that is used.
Comment 1 Michael Morrell 2020-10-13 02:15:22 UTC 
Created attachment 49357 [details]
Source file
Comment 2 Michael Morrell 2020-10-13 02:15:46 UTC 
Created attachment 49358 [details]
Expected output
Comment 3 Michael Morrell 2020-10-13 02:16:07 UTC 
Created attachment 49359 [details]
Incorrect output
Comment 4 Michael Morrell 2020-10-13 02:17:01 UTC 
To run this, just use "gcc -fanalyzer -S -o /dev/null setjmp.c"
Comment 5 David Malcolm 2020-10-13 20:38:31 UTC 
Created attachment 49366 [details]
[PATCH] analyzer: don't use <setjmp.h> in tests [PR97394]

Thanks for filing this bug.  Does this patch fix the testsuite for you?
Comment 6 Michael Morrell 2020-10-14 00:46:34 UTC 
Yes, thanks for the quick fix.
Comment 7 GCC Commits 2020-10-14 20:37:21 UTC 
The master branch has been updated by David Malcolm <dmalcolm@gcc.gnu.org>:

https://gcc.gnu.org/g:974e3975c5bd14ee8817f892532d1e55492227df

commit r11-3894-g974e3975c5bd14ee8817f892532d1e55492227df
Author: David Malcolm <dmalcolm@redhat.com>
Date:   Tue Oct 13 15:56:45 2020 -0400

    analyzer: don't use <setjmp.h> in tests [PR97394]
    
    PR analyzer/97394 reports issues with analyzer setjmp results
    when testing against MUSL.  This patch fixes up gcc.dg/analyzer
    so that it doesn't use <setjmp.h>.
    
    gcc/testsuite/ChangeLog:
            PR analyzer/97394
            * gcc.dg/analyzer/setjmp-pr93378.c: Use test-setjmp.h rather than
            <setjmp.h>.
            * gcc.dg/analyzer/sigsetjmp-5.c: Likewise.
            * gcc.dg/analyzer/sigsetjmp-6.c: Likewise.
            * gcc.dg/analyzer/test-setjmp.h: Don't include <setjmp.h>.
            Provide decls of jmp_buf, sigjmp_buf, setjmp, sigsetjmp,
            longjmp, and siglongjmp.
Comment 8 David Malcolm 2020-10-14 20:55:22 UTC 
(In reply to Michael Morrell from comment #6)
> Yes, thanks for the quick fix.

Thanks for confirming.