Re: [PATCH] testsuite/gcc.dg/compat changes for debugging output

[Eric Botcazou <ebotcazou at libertysurf dot fr>]

> I suspect that the test is aborting early and the output isn't being
> flushed, although on a GNU/Linux system I had a test that was getting
> a segfault but I still got the output.  You could run it with a debugger
> to see if it's aborting via the call to abort(), or earlier with some
> other problem.

Here's the frame:

(gdb) bt
#0  0xffffffff7f100538 in memcpy ()
   from /usr/platform/SUNW,Ultra-60/lib/sparcv9/libc_psr.so.1
#1  0x00000001000655c4 in testvaSld2 (n=8) at struct-by-value-7_y.c:20
#2  0x000000010000d7ec in testitSld2 () at struct-by-value-7_x.c:11
#3  0x0000000100060f54 in struct_by_value_7_x () at struct-by-value-7_x.c:35
#4  0x0000000100073c1c in main () at struct-by-value-7_main.c:12

So it's an unexpected failure.

> What happens when you compile and run the test directly, rather than
> as part of the test suite?

It's better, here's the ouput I get:

poog% ./struct-by-value-7
Sld1 init: ................
Sld1 test: ................
Sld1 testva:
.
..
...
....
.....
......
.......
........
.........
..........
...........
............
.............
..............
...............
................
Sld1 test2: ................
Sld2 init: ................
Sld2 test: ................
Sld2 testva:
F
FF
FFF
FFFF
FFFFF
FFFFFF
FFFFFFF
zsh: segmentation fault (core dumped)  ./struct-by-value-7

> Is there a difference between running it in the background with the output
> redirected and running it in the foreground without redirecting the output?

Definitely. Issuing ./struct-by-value-7 > stv7.log gives an empty log file.

> If the output isn't being flushed when the test aborts early, does the
> following patch fix it?

Yes it does, see the attached log file. The only (very minor) remaining 
complaint is for the missing EOLs at the end of the tests.

Thanks!

-- 
Eric Botcazou

Attachment: gcc.log.gz
Description: GNU Zip compressed data