RFR: 8286429: jpackageapplauncher build fails intermittently in Tier[45]

[Magnus Ihse Bursie]

On Mon, 9 May 2022 23:18:47 GMT, Erik Joelsson <erikj at openjdk.org> wrote:

> The way LauncherCommon.gmk is currently written, it's only meant to be included from "make/module/<module>/Launcher.gmk", or at least only from one single place for each module. This is because the man page generation that happens in LauncherCommon.gmk is module global. For the jdk.jpackage module, LauncherCommon.gmk is now called from two separate sub makefiles, both make/module/jdk.jpackage/Lib.gmk and make/module/jdk.jpackage/Launcher.gmk. These files are called from the top level targets jdk.jpackage-libs and jdk.jpackage-launchers respectively. These top level targets are assumed to be able to run independently of each other, which is normally fine, but now they define the same rules for the same files. This creates a race where one may be deleting files that the other one is creating, causing directories to disappear while files are being written to them. This can fail the build, but also risks silently corrupting the build.
> 
> This patch fixes this by adding a conditional around the man page generation, which helps guarantee that man pages are only processed when called from make/module/<module>/Launcher.gmk. It's a bit of a hack, but it's building on top of the existing design of piggybacking man page generation on the launchers build.
> 
> Also fixing broken whitespace further down in the file (tabs->space).

Marked as reviewed by ihse (Reviewer).

I think this looks good. I'm just curious why this started to show up now. There have been no changes in this area lately, have it? Is it just pure (bad) luck that the race did not happen before?

-------------

PR: https://git.openjdk.java.net/jdk/pull/8618