Re: [gentoo-portage-dev] [PATCH] Make manifest-required-hashes configurable

[Zac Medico]

On 11/06/2017 08:47 AM, Michał Górny wrote:
> W dniu pon, 06.11.2017 o godzinie 08∶44 -0800, użytkownik Zac Medico
> napisał:
>> On 11/06/2017 07:27 AM, Michał Górny wrote:
>>> @@ -148,15 +147,21 @@ class Manifest(object):
>>>  		self.pkgdir = _unicode_decode(pkgdir).rstrip(os.sep) + os.sep
>>>  		self.fhashdict = {}
>>>  		self.hashes = set()
>>> +		self.required_hashes = set()
>>>
>>>  		if hashes is None:
>>>  			hashes = MANIFEST2_HASH_DEFAULTS
>>> +		if required_hashes is None:
>>> +			required_hashes = hashes
>>>
>>>  		self.hashes.update(hashes)
>>>  		self.hashes.difference_update(hashname for hashname in \
>>>  			list(self.hashes) if hashname not in get_valid_checksum_keys())
>>>  		self.hashes.add("size")
>>> -		self.hashes.add(MANIFEST2_REQUIRED_HASH)
>>> +
>>> +		self.required_hashes.update(required_hashes)
>>> +		self.required_hashes.intersection_update(self.hashes)
>>> +
>>
>> Requiring that presence of digests for all manifest-hashes can be
>> problematic, because when we extend manifest-hashes it means that nobody
>> can regenerate a particular manifest without fetching *all* distfiles
>> referenced in the Manifest. It's much more friendly if we generate the
>> missing digests when the distfiles happen to be available in DISTDIR,
>> which is how the portage currently behaves.
>
> This doesn't change. The intersection only strips away hashes that are
> not in 'self.hashes'.

Then the bug must have been introduced by changes to behavior of
digestgen in the same patch. Try adding MD5 to manifest-hashes, and then
try generating a manifest with no internet connection and DISTDIR set to
an empty directory. You'll see that it fails with the patch, but
succeeds with the current master branch.
-- 
Thanks,
Zac