Skillmon

Well, with the current sorting code you'd need to implement such a function yourself. The function would have to be implemented engine dependent, you'd have to implement or use something...

But isn't the level of `xparse` and `l3keys` totally different (at least at the moment)? `xparse` is to create a document interface possibly using only 2e-interfaces, `l3keys` is to create...

Also, what's the problem with using an interface now that already is there, and later add another interface that is possibly accessible from 2e? No matter what the future of...

@FrankMittelbach which of the changes would be ok? E.g., I'd argue changing `\use:x` to `\use:e` doesn't affect structure, but `\use:e` is faster than `\use:x`. Using `\tl_set:No #1 { \l__keys_unused_clist }`...

@FrankMittelbach I really don't get how `\cs_set_eq:NN \l_tmpa_tl \l_tmpa_clist` is assuming more implementation details than `\tl_set:No \l_tmpa_tl { \l_tmpa_clist }` or `\tl_set:Nx \l_tmpa_tl { \exp_not:o \l_tmpa_clist }`. Both assume that...

**Finally benchmarked!** (at least a bit) Benchmarking done for `\keys_define:nn { test } { key .tl_set:N = \l_test_key_tl }` and `\keys_set:nn { test } { key = val }` with...

> which suggests to me that he tl mutator function \tl_set_eq:NN would be the best command to use here. Which would mean to use `\tl_set_eq:NN` with one `tl` variable and...

I merged main, because `\__kernel_tl_set:Nx` was added in the meantime. I hope I spotted every evil line here.

Yes, that indeed sounds like a good idea (and I haven't thought about side effects when `l3build tag` does more than changing version/date information).

In pdfTeX the results are consistent (20% - 30% speed up for 100 - 10000 replications), in LuaTeX they are indeed slower, the new code takes roughly twice as long...