Re: Differential code coverage between 16 and HEAD

On Mon, 2024-04-15 at 21:35 -0400, Tom Lane wrote:
> It's definitely not OK for the standard test suite to include
> internet access.

The update-unicode target is not run as part of the standard test
suite.

>   Seems like we need to separate "download new
> source files" from "generate the derived files".

I'm not sure that's the right dividing line. There are three-ish steps:

1. Download the Unicode files
2. Generate the derived .h files
3. Run tests

If we stop after 1, then do we check in the Unicode files? If so, then
there's inconsistency between the Unicode files and the .h files, which
doesn't seem like a good idea. If we don't check in the files, then
nobody can skip to step 2, so I don't see the point in separating the
steps.

If we separate out step 3 that makes more sense: we check in the result
after step 2, and anyone can run step 3 without downloading anything.
The only problem with that is the tests I added depend on a recent-
enough version of ICU, so I'm not sure how many people will run it,
anyway.

Andres's complaints seem mainly about code coverage in the standard
test suite for the thin layer of C code above the generated arrays. I
agree: code coverage is a good goal by itself, and having a few more
platforms exercising that C code can't hurt. I think we should just
address that concern directly by spot-checking the results for a few
code points rather than trying to make the exhaustive ICU tests run on
more hosts.

Regards,
Jeff Davis