On 3/11/26 07:07, Baptiste Daroussin wrote:
> On Wed 11 Mar 13:45, Kirill Ponomarev wrote:
>> On 03/11, Yuri wrote:
>>> Many ports fail with fs_violation:
>>> =>> Error: Filesystem touched during build:
>>> extra: var/db/pkg/local.sqlite-shm
>>> extra: var/db/pkg/local.sqlite-wal
>>>
>>>
>>>
>>> It looks like pkg is malfunctioning in poudriere because these files are
>>> only opened by pkg.
>>
>> Add this to your /usr/local/etc/poudriere.conf:
>> LOCAL_MTREE_EXCLUDES="/var/db/pkg"
>>
>> This tells poudriere to ignore changes under /var/db/pkg when checking
>> for filesystem violations after build. The sqlite-shm and sqlite-wal
>> files are WAL-mode artifacts left by pkg and are harmless. I guess
>> that came with pkg 2.6.2, it now uses SQLite WAL mode which leaves
>> behind -shm and -wal files after operations on
>> /var/db/pkg/local.sqlite
> 
> This is not needed all poudriere version available in the ports tree
> are handling this fine, just use a version up to date.
> 
> Best regards,
> Bapt
> 
> 

ampere2's from-scratch rebuilds of main-arm64 in 2026 have taken (only
latest is built for main, not quarterly):

237:11:07
242:53:19
256:16:05

ampere2's successful 2026 incremental builds of main-arm64 in 2026 have
taken:

174:51:18
185:22:02

(There is also the distribution time after a build.)

If the port is upgraded just after such a build starts, it can take the
sum of time from a sequence of builds/distributions for an update to be
available in binary form, depending on the relative timing of the
commit: So a notable time delay.

ampere3 building alternately 135arm64-default (a.k.a. latest) and
135arm64-quarterly can lead to even longer intervals between updating a
specific binary. Another issue is that the port tree vintage used only
updates once per cycle-of-2, not after each prior build.

(ampere4 and ampere5 do not take as long but have the sequence-of-builds
based on relative timing structural properties as well.)


-- 
===
Mark Millard
marklmi at yahoo.com

Reply via email to