Hi, In the last days tried/experimented with tmpfs to speed up the compilation process. Sad thing is that before doing this i don't know/estimate the approximate amount of memory the ebuild/package needs, so i could allocate the size plus some amount (10%) just in case. Or if i already have allocated some (ex. 256 MB) just check if it's enough for a specific ebuild and then use it or not. Think IMO this could be done easily, just before qmerge stage get the size of current work directory plus image/sandbox-dir and write it down somewhere. Still don't know/have idea/ where to write this additional info, portage dababase maybe. If present this info could later be fetched by a tool (for example using genlop or other tool, in best case by portage) and people could even manually (initially), see how much memory (approximately as this depends on used USE-flags, version, etc) they need. Or in a perfect case automagically check and mount/unmount tmpfs with a size les s that of some previously given (by user) maximum. In my tests i've seen nearly 50% less time using tmpfs, if there is enough memory of course ;) Thanks Rumen Reproducible: Always Steps to Reproduce: 1. 2. 3.
CFLAGS and USE could have a huge impact on that value
Hi, Of course it greatly depends, but this info will be user-config-specific (if i don't change my CFLAGS,USE-flags too much). Will use previous version's value. In a slowly changing config the value will be near the truth. Have also see some problem with new packages which are much bigger than previous so that mem-size value may be uncorrect. May be use one more ;-) global USE-flag: 'use-tmpfs' and load it in fstab on boot. Rumen
This won't be done by us. The next major release of portage will support a plugin system to some degree that should be capable of supporting this externally.