convert divcoeff.c to D

[WalterBright]

first one in the back end!

[jacob-carlborg]

There's a Visual Stdio project file as well.

[WalterBright]

I'm having enough trouble getting these makefiles to work.

[WalterBright]

It looks like circleci is using an older dmd to compile. It needs to be upgraded to 2.074.1

[wilzbach]

It looks like circleci is using an older dmd to compile. It needs to be upgraded to 2.074.1

Simply bump it here:

https://github.com/dlang/dmd/blob/master/circleci.sh#L5

[WalterBright]

Thanks! That did the trick on circleci.

[MartinNowak]

As said per mail, neither gdc nor ldc support anything like -mv which would restrict us to dmd to build releases. Looking at ~30% speedup for dmd using a better optimizing compiler, that doesn't seem like a smart move.

[MartinNowak]

Sorry, doesn't look like this is using -mv at all. For what would that be needed?

[MartinNowak]

Some info on why you need to update the required compiler would be helpful!
Also need to bump the auto_bootstrap version in posix.mak.

[wilzbach]

How about?

$(DMDSRC backend/_divcoeff.d)

[andralex]

yes please

[ibuclaw]

My guess is that some compilers (or ancient perl wrappers) don't understand -betterC.

[MartinNowak]

My guess is that some compilers (or ancient perl wrappers) don't understand -betterC.

Ah yes, that's another problem. Why do we need -betterC for the backend? Is this some dmc tribute?

[ibuclaw]

Ah yes, that's another problem. Why do we need -betterC for the backend? Is this some dmc tribute?

Well, it looks like -betterC is the reason why jenkins fails.

https://ci.dlang.io/blue/organizations/jenkins/dlang-org%2Fdmd/detail/PR-6907/9/pipeline#step-74-log-128

You could workaround by building with -release, but I already am not a fan of the travisci tests doing that.

[WalterBright]

Some info on why you need to update the required compiler would be helpful!

The older compiler generated code that was looking for the module assert() function in -betterC mode, which had been removed.

-mv

I picked this module because it didn't require -mv

-betterC

The backend doesn't use any D features that require druntime, so I wanted to dogfood and get -betterC working better for such applications.

jenkins failures

This is the same thing that was causing the circleci failures. The solution is to move to a newer D compiler.

[WalterBright]

Essentially, it's time to move forward with converting the backend to D. To that end, gdc, ldc, and the build process needs to support:

1. -mv switch
2. removal of the generated assert helper functions
3. fix Issue 17499 - with -betterC switch, call C's assert failure function rather than druntime's #6901

None of these are tricky or difficult. To that end, divcoeff.d is a trial run to see where the issues are in the toolchain for doing this. Looks like it's doing its job :-)

[ibuclaw]

#6901

Making compilers support features that have not made it to stable (or even in a release) is probably asking for a bit too much.

[WalterBright]

Making compilers support features that have not made it to stable (or even in a release) is probably asking for a bit too much.

These changes are simple and not controversial. Let's move forward with them.

[andralex]

OK I'll pull the switch on this to get in once it passes tests.

[wilzbach]

CC @CyberShadow for updating DAutoTest

[CyberShadow]

@WalterBright Just so we're on the same page, please describe the procedure for building DMD master (incl. this pull request) from scratch without a host DMD compiler (C++ only).

[MartinNowak]

Didn't we just agree that 30% speedup from building dmd with gdc is worthwhile for an upcoming release? This would block this option for quite a while. Also jumping ahead with using latest edge features sets quite a bad precedence for dmd/gdc/ldc compatibility.
Most important though, it's still unclear why we even need -betterC here.

[ibuclaw]

Consider that the oldest compiler building DMD IIRC is using v2.066 of the front-end. I think it's safe to do a leap to 2.068, possibly 2.070 at latest, but I don't really want to break builds when using a release that is not strictly that old.

[wilzbach]

Regarding the current failure: I think you need to add G_DOBJS to the new unittest target in the posix.mak or probably better: add it to D_SRCS.

[JackStouffer]

Walter, the bot comment is going to keep coming back every time there's a new commit. If it doesn't see it's own comment to update, it makes a new one.

[WalterBright]

If it doesn't see it's own comment to update, it makes a new one.

That's not working, then, because I'd delete 10 of them in a row, not one at a time.

[andralex]

so... pull?

[wilzbach]

so... pull?

AFAICT we are still waiting for GDC to catch up or someone to fork the gdmd script and make -betterC a noop:

Wouldn't want to give a precise date, but gcc-8 is first this would land in. As @MartinNowak said, you are free to make -betterC a noop for the dmd-script wrapper.

[WalterBright]

Rebooted as #7714 because I could not get git rebase to work with renamed files (it kept insisting I had not merged them).

[wilzbach]

I could not get git rebase to work with renamed files (it kept insisting I had not merged them).

Not sure what you did, but it works fine for me:

> go pr/6907
> git rebase upstream/master
First, rewinding head to replay your work on top of it...
Applying: convert divcoeff.c to D
Using index info to reconstruct a base tree...
A	circleci.sh
A	src/ddmd/backend/cod2.c
A	src/ddmd/backend/divcoeff.c
M	src/posix.mak
M	src/win32.mak
Falling back to patching base and 3-way merge...
Auto-merging src/win32.mak
Auto-merging src/posix.mak
CONFLICT (content): Merge conflict in src/posix.mak
CONFLICT (rename/rename): Rename "src/ddmd/backend/divcoeff.c"->"src/dmd/backend/divcoeff.c" in branch "HEAD" rename "src/ddmd/backend/divcoeff.c"->"src/ddmd/backend/divcoeff.d" in "convert divcoeff.c to D"
Auto-merging src/dmd/backend/cod2.c
Auto-merging .circleci/run.sh
error: Failed to merge in the changes.
Patch failed at 0001 convert divcoeff.c to D
The copy of the patch that failed is found in: .git/rebase-apply/patch

Resolve all conflicts manually, mark them as resolved with
"git add/rm <conflicted_files>", then run "git rebase --continue".
You can instead skip this commit: run "git rebase --skip".
To abort and get back to the state before "git rebase", run "git rebase --abort".

> git mergetool

Merging:
src/ddmd/backend/divcoeff.c
src/ddmd/backend/divcoeff.d
src/dmd/backend/divcoeff.c
src/posix.mak

Deleted merge conflict for 'src/ddmd/backend/divcoeff.c':
  {local}: deleted
  {remote}: deleted
Use (m)odified or (d)eleted file, or (a)bort? d

Deleted merge conflict for 'src/ddmd/backend/divcoeff.d':
  {local}: deleted
  {remote}: created file
Use (c)reated or (d)eleted file, or (a)bort? c

Deleted merge conflict for 'src/dmd/backend/divcoeff.c':
  {local}: created file
  {remote}: deleted
Use (c)reated or (d)eleted file, or (a)bort? d

Normal merge conflict for 'src/posix.mak':
  {local}: modified file
  {remote}: modified file
Hit return to start merge resolution tool (kdiff3):