Re: [PATCH] pahole: Apply CU-level filters early in loading

[Alan Maguire <alan.maguire@oracle.com>]

On 31/07/2024 19:12, Arnaldo Carvalho de Melo wrote:
> On Wed, Jul 31, 2024 at 06:43:51PM +0100, Alan Maguire wrote:
>>
>>
>> On 31/07/2024 14:26, Arnaldo Carvalho de Melo wrote:
>>> On Wed, Jul 31, 2024 at 09:57:25AM +0100, Alan Maguire wrote:
>>>> On 30/07/2024 23:43, Matthew Maurer wrote:
>>>>> Without this, even with `--lang_exclude=rust` set, running on `vmlinux`
>>>>> with `CONFIG_RUST` enabled will lead to errors like:
>>>>> die__process_function: tag not supported 0x2f (template_type_parameter)!
>>>>> because the filtering doesn't happen until finalization, but unsupported
>>>>> tags are reported during loading.
>>>>>
>>>>> As an added bonus, this should speed up processing of large objects with
>>>>> filtered CUs, as their details will no longer be walked.
>>>>>
>>>>
>>>> One question on this; if we are always doing early filtering like this,
>>>> should the explicit cu__filter() call be removed from pahole_stealer()?
>>>
>>> When I saw the introduction of an extra callback to be used inside the
>>> dwarf_loader I thought that it would be used only for this specific
>>> language filtering feature, i.e. a defensive approach at implementing
>>> this to avoid unintended side effects of doing all filtering at that
>>> point, maybe some other feature somehow depends on the cu__filter()
>>> being called where it was so far.
>>>
>>> But then it is being used for all filtering, so it seems just a way to
>>> reduce the patch size...
>>>
>>> So I'd keep the cu->early_cu_filter() but would use it only for the
>>> language filtering feature, wdyt?
>>
>> So if I understand correctly,
>>
>> 	if (languages.exclude)
>> 		conf_load.early_cu_filter = cu__filter;
>>
>> ? Seems reasonable to me. Thanks!
> 
> yeah, you got it. So this new early filtering is done for the feature we
> have at hand and if other uses of cu__filter() already in place need, we
> can transition to this new mechanism that reuses the cu__filter()
> function signature/semantics.
>

Sounds good! Matthew, can you send a v2 with the above change? It should
still cover us for BTF generation in the kernel case since we always
--lang_exclude rust as long as pahole is recent enough.