Robin Leroy comments

Results 189 comments of


                                            Robin Leroy

jsp: half of the build time is cache files

This is tricky. Unless the code changes, the old versions are stable (but not the dev version). But then if the code changes, all bets are off.

jsp: half of the build time is cache files

That is the time spent _loading_ the cache. Generating the cache from the data files is more like 20 s for each version of Unicode.

ICU-23004 Throw some C++20 at utfiterator.h

(Trying to fix the API bug in the return type of codePoints brought up ICU-22995 again, hence the flurry of intltest commits.)

ICU-23004 Throw some C++20 at utfiterator.h

gcc11-cpp20 is failing because it does not have the defect report from P2210R2; I need to tweak the feature test #if.

ICU-23004 Throw some C++20 at utfiterator.h

Note: I will take a detour through ICU-23132, because it is seriously harming my productivity on this code.

ICU-23004 Throw some C++20 at utfiterator.h

It would probably be a good idea to make utfStringCodePoints a range adaptor object rather than a function…

ICU-23004 Throw some C++20 at utfiterator.h

``` // TODO: If I just take a const Range &unitRange and keep the reference in this object, // then some tests fail & others crash. // Should we take...

ICU-23004 Throw some C++20 at utfiterator.h

> It would probably be a good idea to make utfStringCodePoints a range adaptor object rather than a function… Ah, we can only do that in C++23.

ICU-23004 Throw some C++20 at utfiterator.h

> We can't make a thing in C++17 that happens to satisfy the requirements for a range adaptor? I think we need to inherit from `std::ranges::range_adaptor_closure`. But we could have...

ICU-23004 Throw some C++20 at utfiterator.h

Yes, we definitely don’t want to rebase this mess as-is. 0f63cd5f9d81e21e92a06b3b8f91a4d320d330a7 probably wants to be its own PR, too (by the way, it turns out stdcpplatest is how we get...