svn commit: r566827 - in head/ports-mgmt/poudriere-devel: . files

[Jan Beich]

Bryan Drewery <bdrewery at FreeBSD.org> writes:

> On 3/3/2021 6:13 AM, Jan Beich wrote:
>
>> Bryan Drewery <bdrewery at FreeBSD.org> writes:
>> 
>>> Author: bdrewery
>>> Date: Mon Mar  1 05:11:42 2021
>>> New Revision: 566827
>>> URL: https://svnweb.freebsd.org/changeset/ports/566827
>>>
>>> Log:
>>>   - Update to 3.3.0-540-g9f9007eb8
>> [...]
>>>   * bulk -af: Delete unqueued packages like pkgclean.
>> [...]
>> 
>> I've locally reverted https://github.com/freebsd/poudriere/commit/b19ef339b1a3
>> 
>> Being a ports/ committer I sometimes have to test not just changes to a
>> specific port but also its consumers (from a few to hundreds). Each such
>> list rarely fits on the command line, so I save it into a file and pass
>> to -f. The new behavior of deleting unqueued packages produces churn
>> with intermediate dependencies if not shared by all lists e.g.,
>> 
>> $ git grep -l :devel/py-apscheduler | sed s,/Makefile,, >/tmp/apscheduler.consumers.list
>> $ poudriere bulk -rtkj 122amd64 -f /tmp/apscheduler.consumers.list
>> [...]
>> [00:00:03] Deleting gcc9-9.3.0_1.txz: no longer needed
>> [...]
>> [00:00:04] Deleting llvm11-11.0.1.txz: no longer needed
>> [00:00:04] Deleting llvm80-8.0.1_4.txz: no longer needed
>> [00:00:04] Deleting llvm90-9.0.1_3.txz: no longer needed
>> [...]
>> [00:00:04] Deleting qt5-webengine-5.15.2_1.txz: no longer needed
>> [...]
>> [00:00:04] Deleting rust-1.50.0.txz: no longer needed
>> [...]
>> 
>
> Are you also passing -t or other testing flags? I think it could make
> sense to skip this step for -t at least.

When testing consumers a bustage almost never depends on -t. For ports with
hundreds of consumers enabling -t may produce lots of distracting noise from
ports in poor shape, often unmaintained. For example, I remember disabling
-t when locally testing devel/boost-libs consumers.

Nowadays, I always pass -r and -t (via poudriere.conf) to get more useful
logs and catch unrelated issues. While investigating subtle issues like
unused dependencies or runtime regressions I've often found myself needing
the extra information like NEEDED lines months later.

Also, having packages from different poudriere runs built with and without
"-f list" is useful for the occasional "in vitro" runtime testing inside
jail of various release/arch tuples e.g., when users report bugs.