(a) the unabbreviated form of PSTL

This is a bit ambiguous. Do you mean "Parallel STL"? If yes, I'm not sure it results in less confusion, since it's mostly referred to as the PSTL. The subdirectory is also called pstl.

I think the third line in the commit message should be

Would be nice to include

thanks for the quick response

I note that building on Ubuntu, I am now seeing failures like:

Nice quality-of-life improvement!

I did this (for LibreOffice), and final seems to be only a win when we do LTO and we can de-virtualise a bunch of stuff.

In D102486#2759417, @simon.giesecke wrote:

GSYM conversion still spends a significant time for locking in GsymCreator::insertString even with only 8 threads, so lock contention seems to be generally an issue

According to "graph_header_deps.py", this also removes <compare>'s dependencies on everything-except-type-traits, and drops its cumulative cost from ~40KLOC to ~8KLOC.

random idea from a bystander:

Just a drive-by comment : Possibly the more general form of this needs to be optimised too?

But now you have a "thundering herd" problem?

As a LibreOffice committer, strlen and friends are definitely hot :-)

In D76572#1936861, @Quuxplusone wrote:

Nice. Does LibreOffice have anything (either in clang-tidy or in a paper guideline) against T(x)-style casts? E.g.

Libreoffice has a similar clang-tidy-style cast-checker here:

Nice. Makes me wonder what other loops in clang would work better backwards :-)

Just a drive-by suggestion: Logging the pointer value of the relevant root object in the callees would provide similar benefit to anyone needing to do log tracing

Hah, nice catch, this sounds like something that might make a nice clang-tidy check?

Thanks a lot for this checker - would it be possible to enhance it to also update stuff in associated header files?

You might want to look at the equivalent LibreOffice checkers, which we have been using for some time

Perhaps there should be an "= delete" constructor to make this a compile error?

haha, nice catch.

Nice! I went with a more brute force approach, and I also struggled with logical vs physical const-correctness

There is no need to wrap SVG in HTML if you want to display it in a web-browser, you can just open the web-browser explicitly, they already support opening SVG, use something like

Maybe make AddMap take a lock_guard parameter as recommended here

https://stackoverflow.com/questions/28825749/ensuring-that-current-thread-holds-a-lock-on-a-c11-mutex

?

Just a comment from a bystander:

from someone who is just an observer: What happens when we see IsSExt types

A
B
A

?
Then the last one will re-insert A, which is probably not what you want.

In D40074#926085, @kristof.beyls wrote:

If this is the first canonicalization (of potentially many to come) in the IRTranslator, would it make sense to try and document the canonical representation(s) somewhere and/or somehow?

Sorry, I'm not an LLVM reviewer of any note, can't make any useful decisions

I'm curious if you know why emplace_back is so much faster than resize(size+1) ?

if anyone is interested, I wrote a clang plugin for the LibreOffice codebase to find methods that can return a const ref.

would it be better to internally treat attribute((non_null)) like diagnose_if ? then you get all the warnings for free?

wow, that is neat.
I wonder if it would be legal to teach the compiler to do this if both sides of the expression were enum values of the same type (or possibly more legal if they were scoped enums) ?

Do the use-sites need to be updated? Won't the compiler already simplify such expressions (just concerned that the use-sites may be come less legible)

You might like to take a look at the ConstCharArrayDetector stuff in the LibreOffice codebase here:

Seeing this, you might be interested in the enum/bitset library we use in LibreOffice: