On 2026/03/31 13:41, Claudio Jeker wrote:
> On Thu, Mar 26, 2026 at 11:32:55PM +0100, Alvar Penning wrote:
> > On Wed Mar 25, 2026 at 9:13 PM CET, Claudio Jeker wrote:
> > > Diff below should be better.
> > 
> > Thanks a lot for this diff! I have tested it and deployed it on my
> > Prometheus metrics server. The Go diff looks good on first glance.
> > 
> > Since I had an unfinished prototype of the same update lying around for
> > months, I took the opportunity to add something to your diff. My diff on
> > top of your diff is attached gzipped, as the new distinfo and
> > modules.inc files are ridiculously large.
> > 
> > In a nutshell, there are the following changes:
> > 
> > - Getting rid of the vendored archive, which is hosted on a server of
> >   yours by replacing it with go-module(5). Now, a modules.inc file lists
> >   all the Go dependencies. Got the inspiration from databases/influxdb.
> > - Use Bash as the shell for scripts/compress_assets.sh, as it uses
> >   Bashisms and reports warnings during runtime otherwise.
> > 
> > I am now running my patched version of your diff on my Prometheus server
> > and have not experienced any issues so far. Please consider my change,
> > as itmakes it making it easier for other people to upgrade Prometheus as
> > there is no longer another host involved for the distfiles.
> 
> I will look into that once my update is in. It may be a good idea to
> switch to go-module(5) since the vendor bits are indeed a pain. It takes
> me multiple tries to generate a good file.

basically ok for your update, except scripts/compress_assets.sh does
definitely need either BUILD_DEPENDS on shells/bash or additional
patching, there's a "find ...  -exec bash ... bash" in there.

(the &> is noisy and would be cleaned either by switching to bash or
patching, but GZIP_OPTS does get set how it should be)

I agree with doing the update separately from switching vendor bits.
(the disadvantage with the modules stuff is you can't patch code in
pulled-in modules, only things in the main distfile).