Makefile: add tinygo-test-wasi; like tinygo-test but with -target wasi

[dkegel-fastly]

Skips the standard library modules whose tests don't pass yet in wasi (see recently filed bugs).

[dgryski]

Also dgryski/tinygo-test-corpus#1

[aykevl]

Looks reasonable to me. But also see #2031.

[dgryski]

Also if we're going to start splitting out stdlib package tests, another nice section might be "Packages which pass but fail in CI due to CPU or memory requirements." So far this includes: sort (needs -v to pass until t.Skip is implemented), compress/lzw, and regexp/syntax.

[dkegel-fastly]

Looks like crypto/ed25519 go/format net are also passing?

[aykevl]

Also if we're going to start splitting out stdlib package tests, another nice section might be "Packages which pass but fail in CI due to CPU or memory requirements." So far this includes: sort (needs -v to pass until t.Skip is implemented), compress/lzw, and regexp/syntax.

Possible, although I'd like to fix those issues instead ;)

[aykevl]

Looks like crypto/ed25519 go/format net are also passing?

The net package passes for me, but the other two don't:

$ tinygo test crypto/ed25519
ld.lld-11: error: undefined symbol: crypto/ed25519/internal/edwards25519/field.feSquare
>>> referenced by fe.go:305 (/usr/local/go1.17/src/crypto/ed25519/internal/edwards25519/field/fe.go:305)
>>>               /tmp/tinygo3886713202/main.o:((*crypto/ed25519/internal/edwards25519/field.Element).Square)

ld.lld-11: error: undefined symbol: crypto/ed25519/internal/edwards25519/field.feMul
>>> referenced by fe.go:299 (/usr/local/go1.17/src/crypto/ed25519/internal/edwards25519/field/fe.go:299)
>>>               /tmp/tinygo3886713202/main.o:((*crypto/ed25519/internal/edwards25519/field.Element).Multiply)

ld.lld-11: error: undefined symbol: syscall.RawSyscall
>>> referenced by zsyscall_linux_amd64.go:1661 (/usr/local/go1.17/src/syscall/zsyscall_linux_amd64.go:1661)
>>>               /tmp/tinygo3886713202/main.o:(os.Pipe)
error: failed to link /tmp/tinygo3886713202/main: exit status 1

$ tinygo test go/format
panic: go/parser internal error: unbalanced scopes
FAIL	go/format	0.002s
FAIL

Really hoping for go/format support because it would be a nice feature for play.tinygo.org: formatting code directly in the browser without a server roundtrip.

[aykevl]

Ah, I found one: regexp/syntax. And the culprit is mkCharClass.

[aykevl]

I've profiled it a bit and it looks like the main culprit is heap allocation. I couldn't quickly find a single obvious performance bottleneck (apart from mem.store which takes up 5.26s out of ~30s - too big, but not an obvious bottleneck).
I think it might be possible to make the run function mostly free of heap allocations, at least in all the hot code paths. But that will require some significant refactoring of the interp package.

[dgryski]

compress/zlib should pass now.

[dgryski]

compress/bzip2 should pass once #2387 has landed.

[dgryski]

reflect should be back to passing on wasi now that #2395 is in.

[dgryski]

debug/macho failed for the same reason bzip2 did, and will pass once #2387 is in.

[deadprogram]

Both #2387 and also #2395 have been merged.

Looks like this PR here needs merge conflict resolution/rebase, please.

[deadprogram]

Thanks everyone for working on this! Now merging.