https://reviews.llvm.org/D79773#2131680 has something to do with your .clang-format file, the base LLVM style seems fine.

We may not be consistent across all of LLVM

Thank you @JohelEGP please keep them coming, I'll try and update the patch as soon as I have some fixes, I kind of feel like we are addressing what would have just been downstream bugs anyway. (some are quite bizarre!) so lets squish as many of them as we can before we land this.

Addressed review comments
Added one or two more tests

Still a work in progress but covers I think the case @JohelEGP found.

I'd be interested to know if you felt any of those changes were an unwanted surprise

Thank you I see now... super subtle but I'll take a look.

In D79773#2124231, @JohelEGP wrote:

constexpr plane(const plane& other) noexcept(
    (detail::default_allocator_is_nothrow &&
     std::is_nothrow_copy_constructible_v<T>)) requires(std::copyable<T>)
  : plane{to1d(other), other.sz.w}
{
}

Without the parentheses in trailing requires-clauses, all code following the constructor initializer list is further indented.

In D82707#2118796, @njames93 wrote:

In D82707#2118772, @MyDeveloperDay wrote:

The net result is an actual reduction in the overall %

Is that a bad thing?

The net result is an actual reduction in the overall %

Thank you for this, I’m happy with this and can take a look at removing the os.walk I agree it might be better/quicker

closed by commit rGeb50838ba08d: [clang-format] [PR462254] fix indentation of default and break correctly in…

LGTM

I confirm this is still an issue as of v11 (trunk) and it looks like D21279: [clang-format] Fix some issues in clang-format's AlignConsecutive modes was landed even though its not closed. (the code behind D21279 has been changed probably since it was committed, but the tests appear to be inplace for that issue)

I think we should add all the other cases to ensure we don't break them down the road

Are all the cases from the summary contained in the unit tests?

Just to add, I also think although because there are no tests to enforce it the primary reason is for when we see '\n', breaking the stream on endl but this is actually covered by another rule this also in mustBreak()

@sammccall in the example

In D80940#2071247, @curdeius wrote:

LGTM.

For a second, I thought that you could simplify the code by removing this @try condition (and calling the function FormatTokenLexer::tryTransformTryUsageForC() only if isCppOnly, but given that Objective is a superset of C, it's probably safer to keep it the way it's done now.

IMO this *is* actually a good heuristic: << between string literals indicates some intent to break, and when the stream is entirely literals the formatting is good.

Ensure no crash when CurrentToken is nullptr

Address review comments

Out of curiosity, where does "@try" come from?

In D80950#2069733, @curdeius wrote:

The change seems to me technically sound, but I'm not sure of the scope of its effects. There might be users that rely on this behavior. On the other hand, adding an option to keep the old behavior doesn't seem appropriate, and personally I consider the old behavior as a bug.

In D80933#2066648, @thopre wrote:

That was a speedy patch! I can confirm that my non reduced testcase (git diff -U0 23ac16cf9bd4cc0bb434efcf6385baf083a2ff7b^ 23ac16cf9bd4cc0bb434efcf6385baf083a2ff7b | clang-format-diff.py -i -p1) is fixed with this patch. Thanks! I don't feel confident reviewing the code unfortunately.

We don't really have any performance tests, but the following shows some analysis, all builds are release build:

@steveire Thanks for the additional test cases, I'm reworking how I handle the Macro case as I realized that I didn't actually even try to change the case @rsmith came up with in the first place. I made a decision previous that I couldn't handle any case without the *,& or && to determine that wasn't the case

Addressed review comments
Ensure working with python3
Used improvements from @curdeius's version

Change % colors to help those with deuteranomaly
ensure consistent naming
validate for python3 and Windows
fix flake8 issues
switch to multi-line string templates rather than series of write() calls
add date and current commit hash of when the analysis was run

Would something like this be easier to read?

In D80627#2062332, @curdeius wrote:

@MyDeveloperDay , I know it's a strange request, but could you change (or remove) the background color for 100% case.
I'm partially color-blind and having red and green background in the same table is really hard to distinguish. I guess I'm not alone.
I'd suggest using something like light blue, it doesn't need to stand out anyway.

In D80627#2060463, @JakeMerdichAMD wrote:

Great idea on this! I may borrow this idea and make something similar for some migrations I'm working on.

Sanity check: is this going to be run automatically when docs are generated or done offline with results committed? I don't have a preference, either has pros and cons, but I'm confused as to when this script will be run.

I wondered if it was the < and [] causing issues

@MyDeveloperDay , @Typz -- could you please take a look at these two issues and give your feedback? (E) looks pretty bad. I'm not familiar enough with this patch to judge how easy / hard would it be to mitigate these.

That's a rotate!

In D69764#2058801, @curdeius wrote:

One last thought, how about making the config a bit more future-proof and instead of ConstPlacement accept what was discussed some time ago:

QualifierStyle:
   -   const:     Right

and restrict it to const for the moment?
Maybe renaming to QualifierPlacement or something more appropriate.

fake8 the python
add a totals line at the bottom of the table

I'm adding a summary table which shows LLVM only 44% of all LLVM cpp/h files are clang-formatted. ;-(

I'm running through flake8... almost every line is changing ;-) ...let me update before wasting any more of your time.

Minor change for the simpler review comments before refactoring some of the more involved ones

@rsmith, Thank you for your comments, I don't agree, but thank you anyway.

Begin to cover the cases raised as potential code-breaking changes by ignoring potential MACRO usage
add support for more complex const unsigned long long cases (I believe I can simplify this to a more general pattern)

Fix crash whilst rechecking polly

In D50078#2055609, @Typz wrote:

In D50078#2055227, @srj wrote:

And I also think it should be on by default instead of modifying many .clang-format files. So IMHO if we add an option, it should be opt-out.

I can live with it being opt-out. My big concern is that (as it currently stands) our project gets different 'canonical' clang-format results depending on whether people are using LLVM10 vs LLVM11 (and both are common and supported by our project), which is a headache. I just want a way to get consistent, predictable formatting regardless of the clang-format version.

Adding an option would be an issue there, I think: since clang-format is quite strict regarding options, the option would need to exist in both LLVM10 and LLVM11 if you need to support such use-case...

@rsmith, firstly let me thank you for taking the time to review this, I didn't realize i'd invoke such a reaction that the big guns would start showing up.. fundamentally I agree with you, but let me defend my thesis to help explain my reasons why I still feel this has value.

rebase with master

Fix issue when preprocessor #if/#else is present
Rename the config file name to ConstPlacement
change the command line option to be --const-placement
Add Left/Right into the documentation

I really do appreciate the reviews and the comments especially regarding east/west/left/right/wrong ;-), I know there won't be 100% agreement, but I think most people who even considered cloning the LLVM repo know what we mean by East/West. As such I'm going to leave the internal code that way for now. (and the name of the TokenAnalyzer Files)

I think the buildbot might be failing http://lab.llvm.org:8011/builders/clang-ppc64le-rhel/builds/3664

Address nits

Add more tests and check if a ';' follows the final ]]

Handle more complex nested ObjC calls
Rename function to tryParseSimpleAttributes (not supporting full capability as yet)
Use RAII object to reset the TokenPosition
utilize the fact that Objective calls shouldn't end with "]]" I think... this should allow Attributes at least of the form [[identifier::identifier]]
I feel if this isn't a good enough perhaps we also check for the ; after the second ]

Check for null Tok
Fix up comments