[Ltrace-devel] new ARM failures

[Zach Welch]

On 12/07/2010 05:22 PM, Zach Welch wrote:
> On 12/07/2010 01:47 AM, Joe Damato wrote:
>> On Tue, Dec 7, 2010 at 12:09 AM, Zach Welch <zwelch at codesourcery.com> wrote:
>>> Hi all,
>>>
>>> Today, I built and tested the new 'test' branch on ARM, and several new
>>> failures have appeared. Here is the terse output from the test suite:
>>>
>>> ...
>>> Running ./ltrace.main/parameters.exp ...
>>> FAIL: func_double(3.40*, -3.40*).*= -3.40* in
>>> /scratch/zwelch/ltrace/testsuite/ltrace.main/parameters.ltrace for 0
>>> times ,should be 1
>>> FAIL: <... func_call resumed> \"x\", \"y\") in
>>> /scratch/zwelch/ltrace/testsuite/ltrace.main/parameters.ltrace for 0
>>> times ,should be 1
>>> ...
>>> Running ./ltrace.main/system_calls.exp ...
>>> FAIL: SYS_munmap in
>>> /scratch/zwelch/ltrace/testsuite/ltrace.main/system_calls.ltrace for 1
>>> times ,should be 2
>>> ...
>>> Running ./ltrace.minor/libdl-simple.exp ...
>>> FAIL: Fail to trace libdl loaded function!
>>> ...
>>>
>>> The first and last failures are caused by well known patches, while the
>>> system_calls failure will require more investigation on my part.
>>
>> This is interesting because I don't get the same failures.
>>
>> joe at gcc33:~/ltrace$ uname -a
>> Linux gcc33 2.6.31.12-ER1-efikamx #1 Tue Jan 26 00:57:33 CST 2010
>> armv7l GNU/Linux
>>
>> joe at gcc33:~/ltrace$ file ltrace
>> ltrace: ELF 32-bit LSB executable, ARM, version 1 (SYSV), dynamically
>> linked (uses shared libs), for GNU/Linux 2.6.16, not stripped
>>
>> Only failures I get are:
>>
>> FAIL: func_double(3.40*, -3.40*).*= -3.40* in
>> /home/joe/ltrace/testsuite/ltrace.main/parameters.ltrace for 0 times
>> ,should be 1
>> FAIL: <... func_call resumed> \"x\", \"y\") in
>> /home/joe/ltrace/testsuite/ltrace.main/parameters.ltrace for 0 times
>> ,should be 1
>>
>> libdl and syscalls both work for me.
>>
>> Can you ensure that you were building the correct version of the
>> source? Top of the tree I just built has sha:
>> ccb0b06787e9132ddf9586148223c6dc0cc3ca20.
> 
> Yes, that's what I'm using (plus the fresh stack of patches that I sent
> yesterday). I have made several attempts to eliminate the problems
> through rebuilding the package, so the problem is not that simple to
> fix. I haven't made much progress on it today, but I will keep looking
> at these issues.

I just finished bisecting the tree and found the patch that causes the
system_calls.exp test failure. Specifically, the patch that adds libdl
support [1] causes this problem. I am not certain what part of the patch
causes this failure for me (or why you can't reproduce it), but I
thought it worth reporting in case you have some ideas.

[1] commit f0bd98b3e6753d8609a3054a61f2df6f9cdac10a

-- 
Zach Welch
CodeSourcery
zwelch at codesourcery.com
(650) 331-3385 x743