[Differ] Remove ConsoleDiffer, use DefaultDiffer that utilize ColorConsoleDiffFormatter instead to avoid double diffing process

[samsonasik]

@staabm this is the rework effort from the old PR:

• Improve file-diffing performance #3705
• [Differ] Remove DefaultDiffer, use ConsoleDiffer instead to avoid double diffing process #6482

I tested with diff on json format, it keep same, see https://www.diffchecker.com/YtVb6oKF/

The process is:

• Use DefaultDiffer->diff() as default value
• When pass to FileDiff::__construct() on diffConsoleFormatted, it call

$this->colorConsoleDiffFormatter->format($diff)

like this:

        $diff = $shouldShowDiffs ? $this->defaultDiffer->diff($oldContent, $newContent) : '';
        $consoleDiff = $shouldShowDiffs ? $this->colorConsoleDiffFormatter->format($diff) : '';

        // always keep the most recent diff
        return new FileDiff(
            $relativeFilePath,
            $diff,
            $consoleDiff,
            $rectorsWithLineChanges
        );

Then, to keep the string @@ @@, I use regex on ColorConsoleDiffFormatter->diff().

[samsonasik]

All checks have passed 🎉 @TomasVotruba it is ready for review.

[staabm]

my intial motivation for this change was performance.

did you check whether this new implementation is faster then the previous one?

[samsonasik]

Not yet check on a large diff. The check may need a large diff on a single run to see comparison, yes?

[staabm]

IIRC the reproducer is described in rectorphp/rector#7899

[samsonasik]

I tried that, seems 1 second faster:

Before

bin/rector --clear-cache --dry-run  8.33s user 0.09s system 99% cpu 8.441 total

After

bin/rector --clear-cache --dry-run  6.86s user 0.10s system 98% cpu 7.064 total

[samsonasik]

Updated stop loop on found --- Original and +++ New, got new result:

Before

bin/rector --clear-cache --dry-run  8.33s user 0.09s system 99% cpu 8.441 total

After

bin/rector --clear-cache --dry-run  6.93s user 0.09s system 99% cpu 7.033 total

[TomasVotruba]

@samsonasik What are numbers on one if our private client's project? Real projects are better than Rector itself

[samsonasik]

@TomasVotruba yes, the measure is depends on how many files changed, more diffs, more faster.

I tried in our project with 70 files changed, that means there are diffs, it 3 seconds faster.

Before

../rector-src/bin/rector process src --dry-run --clear-cache  796.63s user 21.67s system 652% cpu 2:05.41 total

After

../rector-src/bin/rector process src --dry-run --clear-cache  772.05s user 20.86s system 647% cpu 2:02.38 total

[samsonasik]

@TomasVotruba I've updated with reduce loop, remove array_map, and it seems faster 7 seconds now, applied 70 files changes, tested on our project:

Before 2:05.41

../rector-src/bin/rector process src --dry-run --clear-cache  796.63s user 21.67s system 652% cpu 2:05.41 total

After 1:58.40

../rector-src/bin/rector process src --dry-run --clear-cache  752.03s user 19.20s system 651% cpu 1:58.40 total

[TomasVotruba]

Let's give it a try 👍