| 1 |
On 5/30/21 2:55 PM, Ionen Wolkens wrote: |
| 2 |
> On Sun, May 30, 2021 at 02:42:01PM -0400, Mike wrote: |
| 3 |
>> |
| 4 |
>> |
| 5 |
>> On 5/30/21 2:28 PM, Ionen Wolkens wrote: |
| 6 |
>>> On Sun, May 30, 2021 at 01:29:12PM -0400, mpagano@g.o wrote: |
| 7 |
>>>> From: Mike Pagano <mpagano@g.o> |
| 8 |
>>>> |
| 9 |
>>>> As the purpose of pkg_pretend is to run sanity checks during |
| 10 |
>>>> dependency calculation time, provide the default implementation |
| 11 |
>>>> and perform CONFIG_CHECK within it. |
| 12 |
>>>> |
| 13 |
>>>> See bug #759238 |
| 14 |
>>>> |
| 15 |
>>> [...] |
| 16 |
>>>> |
| 17 |
>>>> +# @FUNCTION: linux-mod_pkg_pretend |
| 18 |
>>>> +# @DESCRIPTION: |
| 19 |
>>>> +# Check the CONFIG_CHECK options |
| 20 |
>>>> +linux-mod_pkg_pretend() { |
| 21 |
>>>> + |
| 22 |
>>>> + debug-print-function ${FUNCNAME} $* |
| 23 |
>>>> + # External modules use kernel symbols (bug #591832, #759238) |
| 24 |
>>>> + CONFIG_CHECK+=" !TRIM_UNUSED_KSYMS" |
| 25 |
>>>> +} |
| 26 |
>>> |
| 27 |
>>> Is this not supposed to actually run the checks? Correct me if I'm |
| 28 |
>>> missing something but it's just setting the value. linux-mod normally |
| 29 |
>>> run the checks through linux-info_pkg_setup and its check_extra_config. |
| 30 |
>> |
| 31 |
>> Thanks for the analysis, Ionen. |
| 32 |
>> The eclass does not actually run the check as it exists today. |
| 33 |
>> Maybe there was a reason for that when it was originally coded? |
| 34 |
> |
| 35 |
> It does indirectly: |
| 36 |
> # External modules use kernel symbols (bug #591832) |
| 37 |
> CONFIG_CHECK+=" !TRIM_UNUSED_KSYMS" |
| 38 |
> |
| 39 |
> linux-info_pkg_setup; |
| 40 |
> ^^^^^ checks ran here |
| 41 |
> |
| 42 |
> Older nvidia-drivers ebuild was kind of nasty with that, it did its |
| 43 |
> own checks then the linux-mod.eclass' linux-info call did as well |
| 44 |
> (duplicate messages), which I now let linux-mod handle. |
| 45 |
> |
| 46 |
>> |
| 47 |
>> |
| 48 |
>>> For nvidia-drivers, this value will also be lost for the pkg_setup test |
| 49 |
>>> (needed to check, say.. gentoo-kernel emerged in-between) because I |
| 50 |
>>> currently set a local CONFIG_CHECK="..." inside pkg_setup() |
| 51 |
>>> (there's also a conditional CONFIG_CHECK, part of why not global) |
| 52 |
>>> |
| 53 |
>>> Some other ebuilds set CONFIG_CHECK in pkg_setup I believe. |
| 54 |
>> |
| 55 |
>> Maybe virtualbox-modules should do it's own CONFIG_CHECK as nvidia-drivers does. |
| 56 |
>> As that package is the impetus of this patch. |
| 57 |
>> |
| 58 |
>>> Not that I can't change this for nvidia, I guess I could set a global |
| 59 |
>>> scope CONFIG_CHECK with !FATAL-only and += the non-fatal ones in |
| 60 |
>>> pkg_setup to avoid message duplication. |
| 61 |
>>> |
| 62 |
>>> Have same concerns as mgorny wrt exported pkg_pretend, plus I'd also |
| 63 |
>>> need to add my own pkg_pretend wrapper to check MODULES_OPTIONAL_USE |
| 64 |
>>> |
| 65 |
>> |
| 66 |
> |
| 67 |
|
| 68 |
Going to rescind this patch from consideration. |
| 69 |
|
| 70 |
Mike |
Archives