Consider replacing ProcessCapture with something else

[sschuberth]

Regarding the getVersion() invocation mentioned by @mnonnenmacher and @sschuberth, I did some testing. I ran a simple test that executes this method in a loop. What I found out is that there is indeed a small memory leak caused by ProcessCapture calling deleteOnExit() on the temporary files it creates to capture the process output. deleteOnExit() causes the name of the file affected to be recorded in a static Set managed by the internal DeleteOnExitHook system class, so that it can be deleted later.

So each usage of the ProcessCapture class adds three file names to this set (for stdout, stderr, and the temporary directory). This could be cured by ensuring manually that the files are cleaned up, e.g. using a try-finally construct. However, my gut feeling is that the amount of memory wasted here is not really significant.

Originally posted by @oheger-bosch in #3091 (comment)

[sschuberth]

Possible alternatives (alphabetic order):

• https://github.com/05nelsonm/kmp-process
• https://github.com/brettwooldridge/NuProcess
• https://github.com/fleipold/jproc
• https://github.com/jnr/jnr-process
• https://github.com/kgit2/kommand
• https://github.com/pgreze/kotlin-process
• https://github.com/zeroturnaround/zt-exec

[fviernau]

Maybe slightly related: Some of the pip commands the analyzer runs may take quite long. The process capture swallows all output during execution so that no progress can be observed at all. The output is only provided after the execution has finished. This makes it IMO harder to debug. Maybe this can be solved together with migrating to something else.

[sschuberth]

Maybe this can be solved together with migrating to something else.

I totally agree that streaming of (parts of) stdout to logs would be very useful!