Re: git: b16dfd4feec6 - main - mail/thunderbird-esr: update to 140.10.0 (ga)

On 4/25/26 2:06 PM, Bryan Drewery wrote:
> On 4/25/26 7:07 AM, Christoph Moench-Tegeder wrote:
>> Cc'ing bdrewery@ for recent changes in poudriere
>>
>> ## Craig Leres (leres@freebsd.org):
>>
>>> (I don't remember ever seeing this happen before...)
>> Now it see it: the top of your log has
>> : ===== env: FETCH_REGET=0 NO_DEPENDS=yes USER=root UID=0 GID=0
>> and FETCH_REGET=0 means that "make checksum" will not try to re-fetch
>> the distfile on checksum failure.
>>
>> That changed recently in poudriere (here bdrewery@ comes in):
>> https://github.com/freebsd/poudriere/commit/20dc590c29e35866065bcbccc60a8d015289488e 
>>
>
I took a second look and see I am wrong. The networking is enabled at 
fetch phase, before fetch, and disabled after checksum phase. I will 
revert the change in -devel and release. Thanks for the report.



>
> That change is adding `FETCH_REGET=0` when networking is disabled. 
> `make checksum` would fail because there is no networking in that 
> phase. That's why it was done, because it was failing anyway.
>
> The only case where `make checksum` would allow networking is if 
> someone set DISALLOW_NETWORKING=no or 
> ALLOW_NETWORKING_PACKAGES=thunderbird*, which the cluster isn't doing. 
> If those are set then `FETCH_REGET=0` is not set and `make checksum` 
> can rerun (which is going to fail given `make fetch` just failed so I 
> would argue checking for allownetworking is wrong too).
>
>
>>
>> It's causing problems on the package builders, e.g.
>> https://pkg-status.freebsd.org/beefy13/data/135i386-quarterly/301f803e259d/logs/thunderbird-esr-140.10.0_1.log 
>>
>>
>> (The gecko ports have relied on the old behaviour for ages.)
>
> I don't see how because networking is disabled in checksum phase since 
> we first deployed poudriere to the cluster.
>
>
>> This also means, that we can't ever deal with re-rolled distfiles - 
>> which
>> would be a real mess.
>
> We wouldn't want the build to use a file that doesn't match the 
> checksum in distinfo. I don't see the problem here.
>
>
>> It's even more of a mess since it goes completely unnoticed by the
>> individual port maintainer - we typically fetch and verify the distfile
>> outside of poudriere and do a "make makesum" to generate distinfo.
>>
>> More generally: if we allow "fetch", why can't we allow re-fetch?
>> Without having the full context for the change linked above, I'd
>> argue for a revert.
>>
>> Regards,
>> Christoph
>>
>
>