On 01/10/2019 20:34, Paul Gevers wrote: > Hi Sébastien, > > On 01-10-2019 17:35, Sébastien Villemot wrote: >> It failed to build for a 3rd time. So the problem is not transient, and >> I could reproduce it on the porterbox. >> >> For a given C++ source file, the generated assembly file is 303Mb >> large. Then the assembler hits the virtual space memory limit (which is >> only 2Gb on mipsel, if I understand correctly). >> >> The problem did not happen with the toolchain that was used to compile >> octave 5.1.0-1 (gcc 8 and binutils 2.31). However, I’m under the >> impression that it’s not really a bug, but rather another manifestation >> of the limitations of 32-bit architectures (as was discussed on -devel@ >> last August). > > I agree. > >> I found two possible workarounds: either dropping -g, or replacing -O2 >> with -O. >> >> I’m tempted to drop -g for that particular file and particular arch. >> Does that seem ok to you? > > As I have only limited understanding in that area, I can't advice you on > that (maybe other release team members can). But for now I would say, go > ahead, you can always revisit it.
Yes, dropping -g or building with -g1 would be fine. You could also build with -Wl,--no-keep-memory. Cheers, Emilio
