Hi Simon, (Please continue Ccing light...@packages.debian.org)
On Sat, Jan 09, 2021 at 12:14:20AM +0000, Simon McVittie wrote: Mit freundlichen Grüßen Dipl.-Inf. Helmut Grohne Research and Development Application Engineering Phone: +49 (371) 24354 0 | Fax: +49 (371) 24354 020 helmut.gro...@intenta.de | https://www.intenta.de Intenta GmbH ∙ Ahornstraße 55 ∙ 09112 Chemnitz, Germany Managing Director: Dr.-Ing. Basel Fardi ∙ VAT/USt-IdNr.: DE 275745394 Commercial register: HRB 26404 Amtsgericht Chemnitz > On Sat, 09 Jan 2021 at 08:22:26 +0900, Charles Plessy wrote: > > I ran /usr/share/lighttpd/create-mime.conf.pl and only had duplicate > > warnings, that do not cause errors (the scripts returns 0). > ... > > In the latest CI log of lighttpd on amd64, there is: > > > > defconfig FAIL stderr: Duplicate mimetype: '.sdf' => > > 'application/vnd.stardivision.math' (already have 'application/vnd.Kinar'), > > merging to 'application/octet-stream' > > By default, autopkgtest considers any output on stderr to be a failure, > so warnings will make this test fail, unless it's marked with > Restrictions: allow-stderr. > > lighttpd maintainers: please consider whether it should be considered to > be a pass or a fail if this particular test has output on stderr but > exits 0? If it should pass, please add Restrictions: allow-stderr. Yes, this is the purpose of the test: It shall ensure that a lighttpd that uses a default configuration starts without any warnings or errors. We've considered adding allow-stderr already, but doing so would also hide other problems that this test is supposed to catch. I note that this change in media-types consitutes a regression. It used to be free of duplicates (which is why the autopkgtest previously passed). Charles' reasoning that this property cannot be hold up for vendor extensions makes sense. For that reason Glenn Strauss has implemented a change where duplicates on vendor types will be silently merged to application/octet-stream. The change is pending an upload. That still leaves the question of why extensions suddenly are case sensitive (another regression of media-types) even though most consumers treat them as case insensitive. Alexandre Duret-Lutz filed #979554 to that end and we're still waiting for a response from Charles. For these reasons, I think that media-types is rightly blocked from migrating to testing by lighttpd's autopkgtest. We strongly suggest that media-types itself adds unit tests to its own build to ensure some level of sanity on the data it provides: * Disregarding the vendor space, extensions should be unique. * There should be no extensions that differ in case only. Helmut