[lex.charset], [lex.ccon], [re.synopt] Tables placed inside multi-page grammars and code blocks

[eisenwave]

The table in [lex.charset] is being dumped into the middle of the grammar in [lex.universal.char].

Then we have another table that's being dumped into the middle of the grammar in [lex.ccon].

The table in [re.synopt] is being dumped into the middle of the code block inside [re.matchflag], which is clearly wrong. It should be kept within the respective subclause.

[eisenwave]

It seems to me like we have a general problem with tables being dumped into the middle of multi-page code blocks and multi-page grammars. Something should prevent tables from being placed there.

It's also shocking that subclauses would not impose a hard limit on table placement. A table should never be able to slip outside the subclause it's defined in, but here it just gets dumped into the next subclause inside some random code block.

[eisenwave]

@tkoeppe @jensmaurer I think this is a really important issue, and hopefully it's not too late to make some editorial fixes to the DIS if it's affected; this is really broken on so many pages.

[tkoeppe]

The DIS has long been published and circulated, so it is too late to make changes to it.

However, the IS (if any) is still in the future.

[jensmaurer]

LaTeX's table placement machinery places floating tables as "separate" things, outside of the normal flow of text. And that normal flow of text may contain grammar or codeblocks, which are just sequences of paragraphs.

While we can manually repair individual occurrences of bad placement (e.g. by forcing a new page), and maybe for [tab:lex.charset.basic], we should do so unconditionally (because that table is very long), I'm not aware of a LaTeX-level placement hint that would avoid what we're seeing here in a generic fashion.

[eisenwave]

While we can manually repair individual occurrences of bad placement (e.g. by forcing a new page), and maybe for [tab:lex.charset.basic], we should do so unconditionally (because that table is very long), I'm not aware of a LaTeX-level placement hint that would avoid what we're seeing here in a generic fashion.

If you expanded \FloatBarrier from \rSec, you should have the guarantee the tables are not placed outside of the subclause they are defined in. But keeping them inside the subclause is perhaps a lesser issue and more subjective. At the very least, a manually placed \FloatBarrier might fix the problem with [tab:lex.charset.basic] without forcing a page break.

Grammars and code blocks should behave like a "float exclusion region" so that in general, tables cannot be placed inside of them. Placing tables in them seems like placing tables inside other tables. A grammar is effectively a <dl> element and is a contiguous block of associative mapping, just like a table. It's not regular text content. No one in their right mind would put an unrelated <table> inside of a <pre> or <dl> element; that's malformed content.

I've asked ChatGPT about it, and the advice it gave me boils down to:

Modify codeblock to temporarily override \@addtocurcol, \@addtonextcol, and \@tryfcolumn, deferring floats to \@deferlist while the block is active and restoring normal float placement afterward.