Re: circular dependencies in ports tree - how to fix?

On 5/14/26 17:03, Mark Millard wrote:
> On 5/14/26 14:18, Stefan Esser wrote:
>> Am 14.05.26 um 15:04 schrieb void:
>>> Hi,
>>>
>>> On a 15.1 poudriere jail with latest ports tree, the following happens:
>>>
>>> [00:05:30] Sanity checking build queue
>>> [00:05:30] Error: /usr/local/share/poudriere/
>>> bulk.sh:pkgqueue_sanity_check:27:Dependency loop detected:
>>> These packages depend on each other: run:libheif-1.21.2_3
>>> run:ffmpeg-8.1.1,1 run:libcaca-0.99.b20_1 build:libcaca-0.99.b20_1
>>> run:imlib2-1.12.6,2 build:imlib2-1.12.6,2
> 
> Looks like this does not happen with ffmpeg-8.1,1 involved: see later.
> 
>>> These packages depend on each other: run:ffmpeg-8.1.1,1
>>> run:libcaca-0.99.b20_1 build:libcaca-0.99.b20_1 run:imlib2-1.12.6,2
>>> run:libheif-1.21.2_3 build:libheif-1.21.2_3
>>> These packages depend on each other: run:libcaca-0.99.b20_1
>>> build:libcaca-0.99.b20_1 run:imlib2-1.12.6,2 run:libheif-1.21.2_3
>>> run:ffmpeg-8.1.1,1
>>> These packages depend on each other: run:libcaca-0.99.b20_1
>>> build:libcaca-0.99.b20_1 run:imlib2-1.12.6,2 run:libheif-1.21.2_3
>>> run:ffmpeg-8.1.1,1 build:ffmpeg-8.1.1,1
>>> [151amd64-default] [2026-05-12_19h03m56s] [crashed] Time: 00:05:19
>>>
>>> How to fix, please? If not fixable in the short term, would this best
>>> be raised
>>> as a PR in ports infrastructure, or against each port?
>>>
>>> For the time being, I'm going to try commenting out each affected port in
>>> the build list.
>>
>> If I understand the issue correctly, the cause is that poudriere treats
>> all dependencies in the same way, it wants to put RUN_DEPENDS into the
>> buld jail
>> before building the port.
>>
>> I'd expect that graphics/libcaca could be built, if the RUN_DEPENDS
>> would not
>> be depended on at this time.
>>
>> The difference between BUILD_DEPENDS vs. LIB_DEPENDS and RUN_DEPENDS
>> appears
>> to be that BUILD_DEPENDS are not recorded as dependencies in the package
>> manifest.
>>
>> The ports system used to require building and installing of run
>> dependencies
>> before installing a port on the system. That was necessary since
>> traditionally
>> (before the pkg database existed) installation of the port had to be the
>> final
>> step due to the way "make" resolves dependencies.
>>
>> IMHO, poudriere should not put RUN_DEPENDS into the build jail at an early
>> stage (it did, last I checked, a few years ago).
>>
>>
> 
> An issue may be that an install of a pkg might run installed code, say
> to validate operation. That would not be a separate pass after all the
> related installs have happened to make potential cross-references work.
> That may be why dependency looping at the pkg level of detail is checked
> for.
> 
> 
> As for the build: run-depends is checked between "build" finishing and
> "stage" starting.

The above was irrelevant, as was my original "supporting" text later
below. The relevant output from my build is actually:

[00:00:02] Sanity checking build queue
[00:00:02] [main-ZNV4-bulk_a-default] [2026-05-14_14h59m17s]
[pkgqueue_sanity_check] Time: 00:00:01
           Queued: 263 Inspected: 0 Ignored: 0 Built: 0 Failed: 0
Skipped: 0 Fetched: 0 Remaining: 263

So the check was before all the builds but simply did not fail the bulk
run: no loop reproted.

The below is from a somewhat older ports tree that
> allowed the build  . . .
> 
> Using poudriere -v so it is more explicit about when things happen:
> 
> # poudriere -v bulk -jmain-ZNV4-bulk_a graphics/libcaca
> . . .
> [01:01:19] [01] [00:00:00] Building   graphics/libcaca | libcaca-0.99.b20_1
> [01:01:19] [01] [00:00:00] Status     graphics/libcaca |
> libcaca-0.99.b20_1: check-sanity
> [01:01:19] [01] [00:00:00] Status     graphics/libcaca |
> libcaca-0.99.b20_1: pkg-depends
> [01:01:19] [01] [00:00:00] Status     graphics/libcaca |
> libcaca-0.99.b20_1: fetch-depends
> [01:01:19] [01] [00:00:00] Status     graphics/libcaca |
> libcaca-0.99.b20_1: fetch
> [01:01:20] [01] [00:00:01] Status     graphics/libcaca |
> libcaca-0.99.b20_1: checksum
> [01:01:20] [01] [00:00:01] Status     graphics/libcaca |
> libcaca-0.99.b20_1: extract-depends
> [01:01:20] [01] [00:00:01] Status     graphics/libcaca |
> libcaca-0.99.b20_1: extract
> [01:01:20] [01] [00:00:01] Status     graphics/libcaca |
> libcaca-0.99.b20_1: patch-depends
> [01:01:20] [01] [00:00:01] Status     graphics/libcaca |
> libcaca-0.99.b20_1: patch
> [01:01:20] [01] [00:00:01] Status     graphics/libcaca |
> libcaca-0.99.b20_1: build-depends
> [01:01:20] [01] [00:00:01] Status     graphics/libcaca |
> libcaca-0.99.b20_1: lib-depends
> [01:01:23] [01] [00:00:04] Status     graphics/libcaca |
> libcaca-0.99.b20_1: configure
> [01:01:25] [01] [00:00:06] Status     graphics/libcaca |
> libcaca-0.99.b20_1: build
> [01:01:28] [01] [00:00:09] Status     graphics/libcaca |
> libcaca-0.99.b20_1: run-depends
> [01:01:28] [01] [00:00:09] Status     graphics/libcaca |
> libcaca-0.99.b20_1: stage
> [01:01:28] [01] [00:00:09] Status     graphics/libcaca |
> libcaca-0.99.b20_1: package
> [01:01:28] [01] [00:00:09] Finished   graphics/libcaca |
> libcaca-0.99.b20_1: Success
> 
> 
> For reference:
> 
> The above matches: run:libcaca-0.99.b20_1 build:libcaca-0.99.b20_1
> 
> . . .
> [01:00:56] [01] [00:02:24] Finished   multimedia/ffmpeg@default |
> ffmpeg-8.1,1: Success
> (not matching run:ffmpeg-8.1.1,1 <<<===== problem source?)
> . . .
> [01:01:10] [01] [00:00:14] Finished   graphics/libheif |
> libheif-1.21.2_3: Success
> (matching run:libheif-1.21.2_3)
> . . .
> [01:01:19] [01] [00:00:08] Finished   graphics/imlib2 | imlib2-1.12.6,2:
> Success
> (matching run:imlib2-1.12.6,2 build:imlib2-1.12.6,2)
> . . .
> 
> 


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