Allow to gradually remove the auto-tester patching

[wilzbach]

The patching done by the auto-tester makes edits to the win64.mak files nearly impossible, see e.g. the famous dlang/phobos#2526 (oldest non-merged PR at Phobos) or more recently dlang/phobos#5940.

Hence, I propose to check for a NO_AUTOTESTER_PATCHING_V${diffVersion}_${repoName} string, s.t. we can gradually remove these patches.
Should it become necessary to introduce patching again in an unlucky future, you can simply bump diffVersion.

[braddr]

This is one option. The underlying issue(s) need to be tackled, and a number can probably be directly addressed. As far as the referenced pr's go, I still object to parts of them but would have to re-review to recall details. Regardless, I'm about to leave town for a week and have a rule to not make changes that aren't absolutely necessary before traveling. So, this will all wait until after returning.

[wilzbach]

This is one option.

Yep. You don't like it? Do you prefer something else?
Also should we do all three repos at once or just try Phobos and add the similar procedure in a follow-up?
(I will need to modify all three win64.mak files soon as we want to move away from the win64.mak files and drop support for DigitalMars make)

As far as the referenced pr's go, I still object to parts of them but would have to re-review to recall details.

Fair enough, but that's independent to this PR (i.e. the process of removing the patching).

Regardless, I'm about to leave town for a week and have a rule to not make changes that aren't absolutely necessary before traveling. So, this will all wait until after returning.

I hope you had pleasant Christmas & holidays!

[wilzbach]

Ping @braddr - so what's your opinion on this? I would love to start moving away from the win32+64.mak files...

[wilzbach]

Ping. If you don't like grabbing for a string. How about a dedicated file or do you have something else in mind?

[wilzbach]

Friendly ping about this ;-)

[andralex]

ping @braddr can you please help us unclog this

[braddr]

Sorry for taking so long to get back to looking at this. Re-reviewing the changes, the problem I have is that it looks like this strictly providing an off switch that it looks for in the win64.mak file for each repo. I'm not sure I see how that facilitates gradual anything. I see how it facilitates ONE step, which might be enough. What am I missing?

As an aside, cut out the hyperbole, it's not almost impossible to edit the make file, as is manifestly obvious by the number of changes it's had over the years.

[wilzbach]

I'm not sure I see how that facilitates gradual anything. I see how it facilitates ONE step, which might be enough. What am I missing?

Well how else would you do it?
I want to move forward and unify win32.mak, win64.mak and posix.mak into one file and that requires that I can make changes which conflict with the diffs you have.
My approach here was: "set a flag which disabled the diff, s.t. I can remove the diff"
I'm more than open to better suggestions.

[wilzbach]

Ping @braddr - so how can we move forward here?

[marler8997]

The simple solution would be to include @braddr's patches in the repo right?

[wilzbach]

The simple solution would be to include @braddr's patches in the repo right?

We can't do that, because it still results in merge conflicts as the patch can't be applied.

Also AFAICT @braddr doesn't want to open 10 pull requests (for each repo x master and stable) whenever he makes changes to the auto-tester infrastructure, so even if we do a collaborative effort to get rid of all these patch files and coordinate it with @braddr, he might still need to add new ones in half a year later
Of course, including the patchs and removing them entirely would be my preferred way too...

[marler8997]

Sounds like we've painted ourselves into a corner :)

[wilzbach]

Sounds like we've painted ourselves into a corner :)

Yes, but I still have hopes that we can find a way to remove the patching by auto-tester.
@braddr it seems like you don't like the idea proposed here. What would be your preferred way of progressing with this?

[rainers]

@braddr see https://github.com/dlang/dmd/pull/7968/files#diff-180360612c6b8c4ed830919bbb4dd459R56 for an example how to build and run the test suite with compiler overrides without patching the makefiles or d_do_test. We can also add targets to the makefiles to simplify that for the auto-tester.

[wilzbach]

We can also add targets to the makefiles to simplify that for the auto-tester.

Yeah and it would be a lot easier to do with if we have a way to disable the patching within the PR.

@braddr it seems like you don't like the idea proposed here. What would be your preferred way of progressing with this?

Friendly ping.

[wilzbach]

@braddr it seems like you don't like the idea proposed here. What would be your preferred way of progressing with this?

Friendly ping.

pong.

[wilzbach]

Another PR that is lingering in the queue due to the patching errors is dlang/dmd#7414