[oe] [meta-java][PATCH 1/3] ca-certificates-java: Fix sysconfdir for -native recipe

[Yevgeny Popovych]

On 09/17/2018 05:39 PM, André Draszik wrote:
> On Fri, 2018-09-07 at 21:10 +0300, Yevgeny Popovych wrote:
>> When ca-certificates-java-native is built, sysconfdir variable will
>> be set to value that includes WORKDIR.
> 
> What is your use-case for a -native version of this?

I'm glad you asked as this should answer some questions..

I am building a package that contains prebuilt ca-certificates
(including what ca-certificates itself build and JKS),
for this package I need -native version of ca-certificates-java.
We have an immutable (readonly) rootfs, so running scripts at runtime is not an option.
With patches I have recently sent it works pretty well.

The bottom line is that there is a use-case for -native version.
Even if you don't have an immutable (readonly) rootfs - you might want to use it
that way because running scripts is quite error prone.

> 
>> Avoid patching source with this value - use sysconfdir_native.
>>
>> Change-Id: I8ac79c3cd5016a8139d9d8c8d58bc2976d0b6fa3
>> Signed-off-by: Yevgeny Popovych <yevgenyp at pointgrab.com>
>> ---
>>  .../ca-certificates-java/ca-certificates-java_20180516.bb        | 9
>> +++++++--
>>  1 file changed, 7 insertions(+), 2 deletions(-)
>>
>> diff --git a/recipes-core/ca-certificates-java/ca-certificates-
>> java_20180516.bb b/recipes-core/ca-certificates-java/ca-certificates-
>> java_20180516.bb
>> index 2db1915..7db5110 100644
>> --- a/recipes-core/ca-certificates-java/ca-certificates-java_20180516.bb
>> +++ b/recipes-core/ca-certificates-java/ca-certificates-java_20180516.bb
>> @@ -41,9 +41,14 @@ do_patch_append () {
>>      bb.build.exec_func('do_fix_sysconfdir', d)
>>  }
>>  
>> +# sysconfdir will include absolute native sysroot path in -native builds,
>> avoid this
>> +# (see 36.24 of 
>> https://www.yoctoproject.org/docs/2.5/mega-manual/mega-manual.html#faq)
>> +SYSCONFDIR_VALUE_class-target = "${sysconfdir}"
>> +SYSCONFDIR_VALUE_class-native = "${sysconfdir_native}"
>> +
>>  do_fix_sysconfdir () {
>> -	sed -e 's|/etc/ssl/certs/java|${sysconfdir}/ssl/certs/java|g' \
>> -	    -i
>> ${S}/src/main/java/org/debian/security/UpdateCertificates.java
>> +    sed -e 's|/etc/ssl/certs/java|${SYSCONFDIR_VALUE}/ssl/certs/java|g' \
>> +        -i ${S}/src/main/java/org/debian/security/UpdateCertificates.java
>>  }
>>  
> 
> No.
> 
> It's probably best to remove the BBCLASSEXTEND statement in this recipe
> instead, as it was never tested, and is unlikely to be needed.

We have a use-case for -native and this series fixes it.

> 
> The whole idea is to make this independent of the build machine paths. If
> you want a -native version of this, you still need to make sure to point
> this into the yocto sysroot. Otherwise this will try to write to the *build
> machine's* /etc/ directory!
> 
> If you really want a -native version, you need to point this into the
> sysroot, and update this dynamically because of sstate usage, too.

Ok, I suppose that ${sysconfdir_native} has mislead you -
it does _not_ resolve to an absolute path into native sysroot.
The value of ${sysconfdir_native} is always "/etc", it does not mangle in any way (that's why I have used it).
So this don't have any issue with sstate or else.


> 
> 
> Cheers,
> Andre'
> 
> 

-- 
Sincerely,
Yevgeny Popovych