On 13/10/2022 09:44, Ihor Radchenko wrote:
> Max Nikulin writes:
> 
>> I am considering \noalign{} instead of \relax. I was never aware of its
>> effect, but accordingly to The TeXbook it should keep TeX in vertical
>> mode without any action due to empty argument. (Actually I surprised
>> that \relax causes any change of internal state besides parser.)
> 
> If \noalign has less side effects compared to \relax, I'd prefer
> \noalign. Can you confirm this?

My understanding of TeX is not solid enough to confirm this. From The 
TeXbook I have an impression that \noalign is added specially for 
table-like blocks to be used immediately after \cr (low level command 
similar to \\). So I would not use it outside of tables however it may 
be safe. I am not aware of problems with \\\noalign{} in tables. The 
exact effect of \relax is not clear to me yet. I do not have an example 
of negative effect of \relax similar to \hline but outside of tables.

I can try to ask for a better suggestion at stackexchange, but I am 
unsure if the question will be noticed by some person from a TeX engine 
or core LaTeX developers or at least with better understanding of TeX 
internals.

>> Unfortunately \noalign{} just as \relax will not allow @@latex:[1cm]@@
>> on the next line, perhaps @@latex:\noalign{\vskip 1cm}@@ is a workaround
>> to introduce additional vertical space.
> 
> What you are talking about appears to be abusing our exporter. We give
> no guarantees about how \\ is going to be exported internally into
> LaTeX. We should have no obligation to keep use-cases like this.

On the other hand LaTeX backend was transparent to such hacks, so the 
change might be breaking to some users. That is why conditionally adding 
\noalign{} or \relax if \\ is not followed by an export snippet may be a 
better solution.

>> I never used \\* or \\*[10pt] variants of the command. Current stable
>> release should has problems when the line next to \\ starts from * that
>> is not bold marker, besides square brackets.
> 
> I feel like I am missing what you are talking about here.

The rabbit hole is deeper than I thought. The \\ command has its star 
counterpart \\* to prevent page break at this point, and both of them 
allow optional [<LENGTH>] argument. In vanilla LaTeX2e space is allowed 
between \\ and *. However \usepackage{amsmath} redefines \\ command. It 
still can recognize following *, but only on the same line. Unsure if it 
is a bug or feature. So before your commit the following is not a 
problem in *default* Org configuration:

Star\\
@@ignore:@@* on the next line.

Star and brackets\\
*[1em] on the next line.

| Star |
| * in the next row. |
|-|
| Star and brackets |
| *[1em] in the next row. |

However it might be if someone changes list of default packages. Adding 
\noalign{} and \relax should fix the issue just for square brackets.

So at least as a temporary fix \noalign{} should be used for tables 
instead of \relax to mitigate negative effect of the recent patch.

P.S. Packages like longtable might bring more surprises.