Revert "Move DiagnosticReporter to errors.d"

[WalterBright]

Reverts #9484

Once this is reverted, #9350 is next.

[dlang-bot]

Thanks for your pull request, @WalterBright!

Bugzilla references

Your PR doesn't reference any Bugzilla issue.

If your PR contains non-trivial changes, please reference a Bugzilla issue or create a manual changelog.

Testing this PR locally

If you don't have a local development environment setup, you can use Digger to test this PR:

dub fetch digger
dub run digger -- build "master + dmd#9507"

[jacob-carlborg]

Please close, it's fine as is.

[andralex]

@jacob-carlborg as a matter of course, if the project lead disagrees with a large refactoring, we need to go with that. Sorry.

[thewilsonator]

This is not how due process goes.

[wilzbach]

Walter: have you ever tried using DMD as a library? We need to be able to collect diagnostics in DMD as a library, just printing them out to stdout doesn't cut it (think e.g. language server), so until there's a clear argument on how one can do this once you'll have undo this work, I strongly urge you to reconsider.

[wilzbach]

No. Not without a discussion.

[wilzbach]

if the project lead disagrees with a large refactoring, we need to go with that. Sorry.

Sure, but why would anyone still bother contributing?
What do you think will happen to core contributors whose work gets reverted/closed without any good argument?
Yup, they get fed up and leave. This has happened many many times in the past.

[thewilsonator]

You should give a good amount of thought to someone saying that who is giving a dconf presentation on "How to Become a D Contributor"...

[dkorpel]

I'm confused. Jacob is doing valuable work to help the sorry state of D static analysis tools, so why is there so much resistance instead of support? It even seems to comply with Walter's values:
"Successfully using pure functions is regarded with particular favor." - DMD style
"No I/O In Low Level Functions" (and instead) "Push I/O Up Call Stack" - Avoiding code smells
And it's done in relatively small steps instead of a monster PR. The only objection I see mentioned is:

It's a great deal of code added that appears to do nothing.

Yes, there is some temporary boilerplate because it's a gradual refactor. And even once it's finished, the line count may be increased. That also happens when changing gotos into nested functions, it doesn't imply worse code.

Is there another diagnostic reporter design we can all agree with, or it the current error handling design considered ideal and not to be tampered with?

[WalterBright]

there's a clear argument on how one can do this once you'll have undo this work

Further rationale given here: #9350

[WalterBright]

"Successfully using pure functions is regarded with particular favor."

Replacing a call to an impure function with a call to an impure interface doesn't make the caller pure.

"Push I/O Up Call Stack"

Replacing a call to a function with one to an interface is not pushing the call up the stack.

[WalterBright]

@wilzbach > there's a clear argument on how one can do this once you'll have undo this work

#9350 (comment)

[jacob-carlborg]

Replacing a call to an impure function with a call to an impure interface doesn't make the caller pure.
Replacing a call to a function with one to an interface is not pushing the call up the stack.

The DiagnosticReporter PRs did not do this. It was the other PR that you closed #9494, that was in the process of accomplishing that for the lexer. Here's the complete PR that accomplishes that: #9480 (no one has made a single comment in that PR). As you can see, the error function of the lexer is now pure: https://github.com/dlang/dmd/pull/9480/files#diff-b24a7d8934fc062aaa3f71e4e5f96081R2297.

[WalterBright]

As you can see, the error function of the lexer is now pure

Yes, and that is good.

[jacob-carlborg]

As you can see, the error function of the lexer is now pure

Yes, and that is good.

@WalterBright But you don't like it anyway? Because you close the PR #9494 that is the first of several in the process of making this possible.

[WalterBright]

But you don't like it anyway?

That's right. It's much too heavy, as "first of several in the process of making this possible" indicates.

[Geod24]

This branch is 5 months old, and almost 1500 commits behind master. If there's still outstanding issues, please submit a new PR for it.