Compile error from backend gcc when generic int type is defined

[demotomohiro]

Following code causes compile error on backend gcc:

type GInt[T] = int

var a = 1
echo -a

Compile log:

f:\temp>nim c -r -f test.nim
Hint: used config file 'f:\project\Nim\config\nim.cfg' [Conf]
Hint: used config file 'f:\project\Nim\config\config.nims' [Conf]
Hint: used config file 'C:\Users\xyzw\AppData\Roaming\nim\nim.cfg' [Conf]
.........................................................
f:\temp\test.nim(1, 6) Hint: 'GInt' is declared but not used [XDeclaredButNotUsed]
CC: stdlib_digitsutils.nim
CC: stdlib_assertions.nim
CC: stdlib_dollars.nim
CC: stdlib_io.nim
CC: stdlib_system.nim
CC: test.nim
f:\temp\nimcache\test\debug\@mtest.nim.c: In function 'NimMainModule':
f:\temp\nimcache\test\debug\@mtest.nim.c:143:41: error: expected ')' before '-' token
  143 |  T1_[0] = dollar___systemZdollars_3(((NI-24)-(a__test_3)));
      |                                      ~  ^
      |                                         )
Error: execution of an external compiler program 'gcc.exe -c  -w -fmax-errors=3 -mno-ms-bitfields -fmax-errors=3   -If:\project\Nim\lib -If:\t
emp -o f:\temp\nimcache\test\debug\@mtest.nim.c.o f:\temp\nimcache\test\debug\@mtest.nim.c' failed with exit code: 1

Following code compiles successfully.
It seems defining generic int type and -a cause this problem.

Without GInt type:

var a = 1
echo -a

Without generics:

type GInt = int

var a = 1
echo -a

Use bool instead of int for GInt[T]:

type GInt[T] = bool

var a = 1
echo -a

Echo a as is:

type GInt[T] = int

var a = 1
echo a

Multiply -1 instead of -a:

type GInt[T] = int

var a = 1
echo a * -1

distinct int type:

type GInt[T] = distinct int

var a = 1
echo -a

Tested on devel Nim:

nim -v
Nim Compiler Version 1.7.1 [Windows: amd64]
Compiled at 2021-10-25
Copyright (c) 2006-2021 by Andreas Rumpf

git hash: 582468da1cc266d0dd922ec5c620713bbc45d02a
active boot switches: -d:release

[Araq]

What the ...? This is the weirdest bug I've ever seen.

[demotomohiro]

I have tested same code on Nim playground.
It compiles successfully on v0.19.6 but similar compile error comes from gcc in newer version of Nim.
https://play.nim-lang.org/#ix=3CRo

[EliteTK]

It took a bit of effort to bisect this but it looks like this was introduced in 745f164

[EliteTK]

So the buggy code is emitted in ccgexprs in unaryArithOverflow in the mUnaryMinusI case. getSize returns a negative value which gets multiplied by 8. Looks like prior to the bisected version computeSizeAux which was used to implement getSize via computeSize returned conf.target.intSize based on typ.kind if typ.size was negative (which appears to be the case for GInt in this case).

I'm still trying to figure out what exactly the right behaviour is here. If anyone has any ideas that would help.