Replace plain TeX primitives with \iow_* l3file functions

[Witiko]

Closes #1.

[Witiko]

Replacing the (plain) TeX primitives with \iow_* expl3 functions (the reverse of Witiko/markdown@09f4b24) is blocked by latex3/latex3#1105. We will need to wait until the fix (latex3/latex3@a98fde5) from 2022-06-26 has trickled down to the texlive/texlive:latest Docker image.

[Witiko]

Dear @josephwright, can you please advise why ce1cb85 would cause the CI to fail? My reading of the l3file documentation is that \iow_open:Nn, \iow_now:Nn, and \iow_close:N should have the same semantics as the \immediate variants of \openout, \write, and \closeout. Yet, replacing the latter with the former leads to failure. I am unsure if this is a bug in l3file or my misapprehension of its semantics.

[josephwright]

Did you allocate first? We use a pool of streams, so you need \iow_new:N before you can open.

[Witiko]

@josephwright I used \g_tmpa_iow and thought it was always allocated.

[Witiko]

@josephwright I tried to zero in on the cause of the issue with the following results:

1. After excluding ConTeXt tests and replacing the \immediate variants of \openout, \write, and \closeout with the \iow_open:Nn, \iow_now:Nn, and \iow_close:N functions from expl3-generic in f38c3c9 and b0c8939, all tests pass.
2. After including the ConTeXt MkIV test in bb08ef0, all tests still pass.
3. After including the ConTeXt MkII test in e694f3c, the MkII test fails.

This leads me to the conclusion that there might be an issue with using l3file from expl3-generic in the ConTeXt MkII format that needs to be fixed. I will try to come up with an MWE after I have returned from the vacation next week.

[josephwright]

Testing on ConTeXt is minimal, and MkII is now two versions 'obsolete' (and hard to set up, as it needs Ruby): I very much doubt there are many real users.

[Witiko]

@josephwright That may well be, but I won't break compatibilty without a reason. MkII may be old but it is a superset of plain and usually well-behaved. Would you be interested in a patch to expl3-generic when I discover the cause?

[Witiko]

@josephwright Mystery solved: It seems that in ConTeXt MkII, \__iow_new:N needs to be patched similarly to how we already patch \__ior_new:N for ConTeXt MkIV, see bc1d9cc. I offer a patch for LaTeX3 in latex3/latex3#1114.

[Witiko]

For my future reference: If latex3/latex3#1114 gets merged, we revert 110c8d4 and bc1d9cc, and squash before merging #3. Otherwise, we revert only 110c8d4, better document bc1d9cc, and squash main..e694f3c before merging #3.

[Witiko]

Try again on Saturday. If successful, revert 82cb941 and merge the PR.