Building the archive index is costly

[gsmet]

Creating this here even if the code lies in SmallRye Common.

Index.of represents 4.54% of the CPU startup time, which is far from negligible. I'm posting one of the call sites here but there are others.

[dmlloyd]

I'm trading init-time work for classload-time speed. That said, if we write the archives in sorted order, it should reduce this a little. That's something we can do in the modular writer if we aren't already. There might also be a couple individual code tweaks to make this faster as well.

[gsmet]

Claude suggested a few changes, which led to some marginal improvements but the sorting and the binary search are quite costly, from what I have seen.

[dmlloyd]

I am skeptical that binary search itself is costly. But perhaps there is something else going on (extra string manipulation, or method handle overhead, or something like that).