In order to make FEATURES=compress-build-logs more scalable for emerge --jobs, we should offload the compression to a separate process (the PipeLogger class will manage this process). Eventually we can make the compression program configurable, but that will required updates to all the code that assumes gzip compression.
I wonder if, in the process of doing this, we should rework how it operates entirely and just compress at the end. I always assumed the value for this was when Portage is configured to save build logs, rather than in PORTAGE_TMPDIR during the build. It makes it a pain to follow progress with e.g. tail -f as-is...
(In reply to Sam James from comment #1) We could possibly write an uncompressed log to ${T}/build.log while we simultaneously write a compressed log to PORTAGE_LOGDIR. It might be better if the log in PORTAGE_LOGDIR is always compressed for FEATURES=compress-build-logs, in case that's on a network share or something.
