[OE-core] [PATCH] glibc: Fix locale DEPENDS

[Joshua Watt]

On 8/28/19 11:09 AM, Richard Purdie wrote:
> On Tue, 2019-08-27 at 07:45 -0500, Joshua Watt wrote:
>> gettext is required to generate the glibc locales in do_compile. If not
>> present, glibc will skip the generation which isn't reproducible.
>>
>> Signed-off-by: Joshua Watt <JPEWhacker at gmail.com>
>> ---
>>   meta/recipes-core/glibc/glibc.inc | 2 +-
>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/meta/recipes-core/glibc/glibc.inc b/meta/recipes-core/glibc/glibc.inc
>> index 252fd56c13c..f1a6ae2a245 100644
>> --- a/meta/recipes-core/glibc/glibc.inc
>> +++ b/meta/recipes-core/glibc/glibc.inc
>> @@ -6,7 +6,7 @@ DEPENDS = "virtual/${TARGET_PREFIX}gcc libgcc-initial linux-libc-headers"
>>   
>>   PROVIDES = "virtual/libc"
>>   PROVIDES += "virtual/libintl virtual/libiconv"
>> -inherit autotools texinfo distro_features_check systemd
>> +inherit autotools texinfo distro_features_check systemd gettext

FYI: I forgot to send this one in-reply-to a previous patch. There is 
more history here: 
http://lists.openembedded.org/pipermail/openembedded-core/2019-June/284158.html

> I suspect this may not do what you expect.
>
> At least as I read the class and recipes, glibc sets
> INHIBIT_DEFAULT_DEPS which means no gettext-native dependency is added,
> instead, configure has --disable-nls added.
Ah, yes I see that also.
>
> Is that what we want?

I guess I'm not sure exactly what we want; It was a while ago so I don't 
remember exactly what glibc was doing, but it looks like it was 
sometimes generating locales (e.g. the first time it built) and 
sometimes not (when rebuilding) depending on the presence of gettext.

>
> I'm curious to understand how glibc locales are ever generated
> correctly and what we're aiming to add here (a gettext-native
> dependency?)

I'm not sure how it ever works correctly :). Perhaps it would be best to 
just drop this patch for now until we take a more serious look at 
reproducible in the presence of partial recipe rebuilds. I'd rather 
focus on getting the existing QA test running ATM.

>
> FWIW gettext-native is a very heavy thing to add in as a dependency
> from a build time perspective.
Yes, that was discussed which is what led to using gettext.bbclass that 
might use gettext-minimal-native, but as you pointed out this isn't 
whats happening.
>
> Cheers,
>
> Richard
>