Support variable declaration in for loop setup

[ChAoSUnItY]

• Resolve Declare variables where needed #115 , this allows variables declare in the scope of for loop body, to avoid loose variable liveness, in other word, variable can have stricter scope liveness by declaring in setup section of for loop.

[jserv]

CI complains below:

tests/check-snapshots.sh
Files /dev/fd/63 and /dev/fd/62 differ
Error: writing output failed: Broken pipe
cat: write error: Broken pipe
Files /dev/fd/63 and /dev/fd/62 differ

@vacantron, can you help resolve?

[vacantron]

CI complains below:

tests/check-snapshots.sh
Files /dev/fd/63 and /dev/fd/62 differ
Error: writing output failed: Broken pipe
cat: write error: Broken pipe
Files /dev/fd/63 and /dev/fd/62 differ

@vacantron, can you help resolve?

The new cfront feature is going to change the CFGs. Use tests/update-snapshots.sh to update the outdated ones.

By the way, the changes causes the inconsistent behavior with GCC build while unwinding the phi instruction in the second stage, and breaks the self-hosting. I'm trying to figure it out.

[vacantron]

Rebase the master branch after #113 being merged, and re-run the tests/update-snapshots.sh . The error should be solved.

[ChAoSUnItY]

That's weird, I updated snapshot after rebase and it still panics on stage 2.

[vacantron]

That's weird, I updated snapshot after rebase and it still panics on stage 2.

The second stage is broken by the SSA now, but the snapshot test should be passed tho.

I run head -c20 tests/snapshots/xxx.json | tail -c19 for the hello and fib, and both of them has 469 basic blocks. It is definitely incorrect. The fib one should have ten more than the hello. Can you try it again with make distclean and make config ARCH=arm ?

[ChAoSUnItY]

Can you try it again with make disclean and make config ARCH=arm ?

After running disclean and config ARCH=arm, now running tests/update-snapshots.sh would stuck forever (I terminted it after running for 2.6 mins). Can you confirm this weird behavior?

[vacantron]

After running disclean and config ARCH=arm, now running tests/update-snapshots.sh would stuck forever (I terminted it after running for 2.6 mins). Can you confirm this weird behavior?

There is a typo in the commands that I provided previously. Try make distclean config ARCH=riscv out/shecc-stage1.elf and tests/update-snapshots.sh again. Do you run them on Linux?

[ChAoSUnItY]

Do you run them on Linux?

Yes, specifically, on Arch Linux.

Try make distclean config ARCH=riscv out/shecc-stage1.elf and tests/update-snapshots.sh again.

It still failed on stage 2.

[vacantron]

Yes, specifically, on Arch Linux.

Can you test the jq command with -S option separately on your machine? The result seems not to be sorted in ascending order by the _gvid.

[ChAoSUnItY]

I can confirm that using jq -S -c . hello.json > hello.json (from tests/update-snapshot.sh) does only sorts entries by json keys but _gvid value.

[jserv]

Rebase the latest master branch and re-generate snapshots.

[vacantron]

Leave a blank line between the subject and the body of the commit message, and use git rebase -i to squash some commits

[jserv]

Leave a blank line between the subject and the body of the commit message, and use git rebase -i to squash some commits

See also: https://cbea.ms/git-commit/

[jserv]

Thank @ChAoSUnItY for contributing!