On Thu, 10 Feb 2022 15:52:50 GMT, Magnus Ihse Bursie <i...@openjdk.org> wrote:
>> Current adhoc version build strings are not ideal. Some of the problems: >> * A build number of "0" is inserted, which make the version string look >> like it's an official build, at least when not reading carefully >> * The version string gives little indication on what source code the build >> was based >> >> Also, an error was discovered in how the build system generates version >> strings without a build number (since this basically never happened before). >> A version without a build number, and without a PRE value, but with an OPT >> value, should have a "+-" separator, according to JEP 223. While this was >> correctly handled, the "+-" separator was also applied to versions without a >> build, but with both PRE and OPT. In this case, the "+" should be omitted, >> according to JEP 223. This bug is fixed by this commit. >> >> Ideally, the adhoc version string should include something along the lines >> of the output of `git describe`. However, since the version string is set at >> configure time, not build time, this will almost immediately become >> misleading. :-( A substitute is proposed in this patch, where the branch >> name is included in the OPT string (if it's not `master`). >> >> Finally, this patch fixes hotspot so it can properly build without a build >> number. This was the last blocker for why the build system always required >> the "0" as build number before. To facilitate interoperability with external >> tools (like jib) that still sets build number to "0" to really signal "no >> build number", a build number exactly matching "0" will be interpreted as >> having no build number. > > Magnus Ihse Bursie has updated the pull request incrementally with two > additional commits since the last revision: > > - Remove "build 0" from GHA builds > - Warn that build 0 is "no build" This was more problematic than expected. Two tests made unfounded assumptions about version string format; one was simple to fix but the other tricky to track down, since I initially could not reproduce it locally but only in our CI testing. (It turned out that this was since the CI testing made a specially formatted version string.) The GHA actions all made use of "--with-version-build=0", and thus all of them failed when trying to upload builds. It took me a while to realize that this was due to file names with version strings in them, and not general shakiness of the GHA builds. I think this is finally ready for review now. :-) ------------- PR: https://git.openjdk.java.net/jdk/pull/6152