Re: [PATCH] Fail if dwz fails.

[Tom de Vries <tdevries@suse.de>]

On 5/16/21 10:14 PM, Mark Wielaard wrote:
> Hi Martin,
> 
> On Fri, May 14, 2021 at 12:46:15PM +0200, Martin Liška wrote:
>> On 5/13/21 4:07 PM, Mark Wielaard wrote:
>>>> On 4/22/21 1:11 PM, Martin Liška wrote:
>>>>> Right now, dwz return code is 0 or a small integer value (<= 3)
>>>>> for situations like:
>>>>>
>>>>> - dwz: Too few files for multifile optimization
>>>>> - dwz: Multi-file optimization not allowed for different pointer sizes or endianity
>>>>>
>>>>> These are not fatal errors and the script should continue. On the other
>>>>> hand abort or segfault error values should cause failure of the script.
>>>>>
>>>>> Let's reserve values 1-16 for a recoverable dwz exit codes.
>>>
>>> With "recoverable" you mean that dwz did do compression, but could not
>>> deliver optimal compression?
>>
>> No, I speak about cases, where compression cannot be done
>> (different endianess, ptr sizes, not beneficial, any other reason).
> 
> I think we are talking about somewhat the same thing. In those cases
> dwz does something. The result is just not optimal.
> 
>>> It would be good to have an ack from the dwz developers that 1 to 16
>>> are "recoverable" errors (I added dwz@sourceware.org to CC).
>>
>> Yes. That's why I CCed Tom.
>>
>>>

AFAIK, the exit codes are either 0 or 1 (excluding abort/assert).

FWIW, also invalid input dwarf results in exit code 1.

>>> Would it be possible to tweak the find-debuginfo.sh script to avoid
>>> them? e.g. Could we see how many arguments we have so that we only use
>>> -m when there are 1+ debug files? And/Or detect debug files using
>>> different endianess and ptr sizes so they are processed in different
>>> batches?
>>
>> No, I would leave it to dwz. There may be other reasons.
> 
> But for the reasons we do know about we could improve
> find-debuginfo.sh to do the right thing (not use -m if there is only
> one file, sort files by endianess/ptrsize). Ideally you call
> find-debuginfo.sh and it does the right thing even if you only have
> one ELF file or a mix of ELF files.
> 
>>> Ideally an error code from dwz means something went terribly wrong and
>>> we abort find-debuginfo.sh. IMHO.
>>
>> I would allow the mentioned "soft" error codes.
> 
> Couldn't optimize is an "soft" error indeed. I don't know about the
> others.

Couldn't optimize is not an error:
...
$ ./dwz hello ; echo $?
0
$ ./dwz hello ; echo $?
./dwz: hello: DWARF compression not beneficial - old size 3372 new size 3372
0
$ ...

OTOH, failing to create a file specified with -o is:
...
$ ./dwz hello -o hello.1; echo $?
0
$ ./dwz hello.1 -o hello.2; echo $?
./dwz: hello.1: DWARF compression not beneficial - old size 3372 new
size 3372
1
$
...

FWIW, I don't have a strong opinion on this patch.

As dwz developer, more bug reports means a better dwz, so this is good
for me.

I just wonder whether this is a good choice for package developers.
Isn't a dwz crash for them just a case of "no debuginfo compression"?

Thanks,
- Tom