Parallel `make check` doesn't run in the right order

[Manishearth]

I ran make check -j7, and some of the tests were failing because the syntax and rustc libs in $target/stage2/lib/rustlib/....../lib/ hadn't been built yet.

I guess the makefile needs some tweaks?

[kmcallister]

I've seen this too. I usually do make -j8 all && make check but it'd be great if this worked.

[Manishearth]

This seems to be specific to the pretty tests. Goes away with make rustc-stage2 && make check-stage2, so I assume a make dep is missing.

[pnkfelix]

cc me

[rprichard]

@pnkfelix AFAICT, the PRETTY_DEPS$(1)_H_$(3)_pretty-... definitions in this commit are not defining the expected variables, because the definitions are outside any function, so $(1) and $(3) expand to nothing.

make -pn shows this line:

PRETTY_DEPS_H__pretty-rpass-full = $$(HLIB$(1)_H_$(3))/stamp.syntax $$(HLIB$(1)_H_$(3))/stamp.rustc

I'll work on a fix.

[rprichard]

@pnkfelix Do you have a preference on how to fix the run-make/simd-ffi test with parallel make check?

I have a fix now, but it's not successfully tested.

Parallel make check now fails because the run-make/simd-ffi test is run in parallel, and it has multiple phony targets that generate the same output file. I think it's easy to fix that test. Alternatively, we could disable parallelism for run-make tests by modifying src/etc/maketest.py somehow: (1) add -j 1 to the command-line or (2) filter/clear MAKEFLAGS. I'm leaning towards -j 1 if we want to disable parallelism. I think we require GNU make, so I would think -j 1 wouldn't break anything.

[rprichard]

I fixed run-make/simd-ffi. I'll leave maketest.py alone for now.