Another precedence issue in math parsing

[mathmil]

Thanks for fixing my last bug report so quickly, this one is quite similar.
typst:

$
  (b)(c)/(d)
  !(a)/(b)
  \ !(a)/(b)
$

typst output:

pandoc latex output, identical for newest nightly build and newest release:

\[\begin{array}{r}
\frac{(b)(c)}{d}\frac{!(a)}{b} \\
!\frac{a}{b}
\end{array}\]

Only the first two cases are wrong, but I find it weird that the second and third one differ.

[jgm]

Yes, this is definitely quite puzzling:

$  
!(a)/b
\
!(a)/b
$

I can't currently see why these two cells would be parsed differently.

[jgm]

Pandoc yields

\frac{!(a)}{b} \\
!\frac{a}{b}

And neither is correct. It should be !\frac{a}{b}.

[jgm]

Related case

$
+(a)/(b)
$

Here pandoc treats the + as part of the numerator, while typst does not.

[jgm]

I think what's happening here is that code intended to get the right grouping for

f(x)/3

is doing the wrong this for

!(x)/3

and

+(x)/3

[jgm]

It's hard to guess the exact rule typst is using. Note that

f(x)/3

has f(x) in the numerator, while

f_1(x)/3

does not have f_1(x) in the numerator.

[jgm]

Ah, but

f_a(x)/3

does have f_a(x) in the numerator. So I guess the rule has something to do with being followed by a letter?

[jgm]

In my opinion there's a bit too much "magic" in typst's parser. It makes it hard to predict.

[jgm]

Interestingly f_alpha(x)/3 and f_"alpha"(x)/3 both pattern with f_a(x)/3.
So it's not quite "preceded by a letter."

[knuesel]

This is actually an issue with the parsing of subscripts. I think the current behavior was introduced in Typst 0.5 to allow typesetting function calls in subscripts/superscripts without extra parentheses. So f_a(x), f_alpha(x) and f_"alpha"(x) are all interpreted as a function call in subscript of f. Then naturally the fraction f_a(x)/3 is parsed such that f_a(x) is the numerator.

Meanwhile 1(x) is not treated as a function call, so f_1(x)/3 is parsed as f_1 (x)/3.

I don't think the exact rules are documented anywhere. I guess you can get close enough by treating as a subscript function call anything that looks like a variable name or "text" followed by parentheses.

(Many people are unhappy with the current rules and there's been a lot of discussions about a redesign but no decision yet as far as I know.)

[jgm]

Many people are unhappy with the current rules and there's been a lot of discussions about a redesign but no decision yet as far as I know.

If you could link to some of those discussions, it would be useful.

[knuesel]

There's some discussion on GitHub in typst/typst#985 and typst/typst#2086 but most of it was on Discord, in particular:

https://discord.com/channels/1054443721975922748/1176475099940007976

https://discord.com/channels/1054443721975922748/1176478139757629563/1297121034209722371

https://discord.com/channels/1054443721975922748/1176478139757629563/1249499031873388616

https://discord.com/channels/1054443721975922748/1262525590175617085/1283107875144601715

[jgm]

Reopening because we still don't get the original case quite right:

document(body: { math.equation(block: true, 
                               body: { math.frac(denom: text(body: [d]), 
                                                 num: { math.lr(body: ({ [(], 
                                                                         text(body: [b]), 
                                                                         [)] })), 
                                                        math.lr(body: ({ [(], 
                                                                         text(body: [c]), 
                                                                         [)] })) }), 
                                       text(body: [!]), 
                                       math.frac(denom: text(body: [b]), 
                                                 num: text(body: [a])), 
                                       linebreak(), 
                                       text(body: [!]), 
                                       math.frac(denom: text(body: [b]), 
                                                 num: text(body: [a])) }, 
                               numbering: none), 
                 parbreak() })

We have \frac{(b)(c)}{d} but it should be (b)\frac{c}{d}.

[jgm]

OK I think I've got it now.