[Buildroot] [PATCH] core/legal-info: use hash file from version sub-dir

[Luca Ceresoli]

Hi Yann, Joshua,

On 14/07/2017 01:32, Joshua Henderson wrote:
> Yann,
> 
> On 07/13/2017 03:19 PM, Yann E. MORIN wrote:
> 
>> diff --git a/package/pkg-utils.mk b/package/pkg-utils.mk
>> index accf48c464..f285395267 100644
>> --- a/package/pkg-utils.mk
>> +++ b/package/pkg-utils.mk
>> @@ -86,9 +86,12 @@ endef
>>  define legal-license-file # pkgname, pkgname-pkgver, pkgdir, filename, file-fullpath, {HOST|TARGET}
>>  	mkdir -p $(LICENSE_FILES_DIR_$(6))/$(2)/$(dir $(4)) && \
>>  	{ \
>> -		support/download/check-hash $(3)/$(1).hash $(5) $(4); \
>> -		ret=$${?}; \
>> -		case $${ret} in (0|3) ;; (*) exit 1;; esac; \
>> +		if [ -d $(3)/$($(PKG)_VERSION) ]; then \
>> +			support/download/check-hash $(3)/$($(PKG)_VERSION)/$(1).hash $(5) $(4); \
>> +		else \
>> +			support/download/check-hash $(3)/$(1).hash $(5) $(4); \
>> +		fi; \
>> +		case $${?} in (0|3) ;; (*) exit 1;; esac; \
>>  	} && \
>>  	cp $(5) $(LICENSE_FILES_DIR_$(6))/$(2)/$(4)
>>  endef
>>
> 
> I think this has unintended side effects.  If there is already a $(PKG)_VERSION directory for patches, 
> it will look for the hash file there even if there are no hash file differences, which results in missing
> license file hash.
> 
> I believe it would work if you test -f on the actual .hash file instead of just the directory.  But, do note
> that if there are multiple license files per package, and only one of them is different, this still results
> in putting all hashes including duplicate ones in separate hash files.

I agree with Josh here.

A cleaner solution would be to put only version-bound hashes in
version-named subdirs, and leave all other hashes in the base directory.
This would remove all duplication and avoid unintended disabling of all
hashes when one <version>/*.hash file is created, which would happen
only for those versions that actually have that file.

I think this can be achieved in a simple way by cat-ing together
$(3)/$(1).hash and $(3)/$($(PKG)_VERSION)/$(1).hash in a temporary file
and feed that to check-hash. Additionally if check-hash were able to
read the hashes from stdin, no temporary file would be even needed, but
this is not necessarily a good idea.

-- 
Luca