(llvm) types and blocks are parsed incorrectly

[gnzlbg]

Describe the issue

The LLVM types and blocks are parsed incorrectly.

Which language seems to have the issue?
LLVM

Are you using highlight or highlightAuto?
No

Sample Code to Reproduce

;; foooo
define i32 @mul_add(i32 %x, i32 %y, i32 %z) {
  entry:
    %tmp = mul i32 %x, %y
    %tmp2 = add i32 %tmp, %z
    %tmp3 = add i32 %tmp, 0
    ret i32 %tmp3
}

Expected behavior
I expect:

• define to be a .hljs-keyword
• i32 to be a .hljs-type (its parsed as a keyword, all types are described here: https://llvm.org/docs/LangRef.html#type-system)
• @mul_add to be a .hljs-title (its a function name)
• all the %... to be .hljs-symbol
• 0 to be a .hljs-literal
• mul, add, ret to be .hljs-built_in or .hljs-keyword (these are parsed as keywords)
• entry: is a block, so i'd expect this to be maybe a .hljs-section ? not sure
• ;; fooo to be a comment

[joshgoebel]

Why is 0 a literal and not a number? literal is typically reserved for built-in literal values in a language, ie: NULL, true, false, etc...

Tagging beginner friendly.

• We don't really match types now at all over than i* (as keyword)... I assume that rule was added to match i32, i8, i64, etc... so it'd be nice to have an explicit types list (for accuracy if nothing else).
• For the label I'd suggest meta I think. In other languages we use symbol but we're using that for "%identifier"... are these symbols or variables? If they are variables we could just use variable and then use symbol for the labels.

Many styles make almost a "subtheme" for Markdown (which is most often where section is used) which can make it a poor choice to general programing languages. We are working on more semantic descriptors for things though but that's a bit of a long-term mission...

Are these truly called "blocks" in LLVM?

[joshgoebel]

I see string are permittable (in the grammar)... is an empty string allowed? ie, ""?

[joshgoebel]

Ping.

[joshgoebel]

Ping.

[allejo]

The closest thing I can find about 0 being a "literal" is when it's preceded by a !...? For example, in llvm's "Module Structure" example snippet.

; Named metadata
!0 = !{i32 42, null, !"string"}
!foo = !{!0}

With PR #2830, I see the following difference in our highlight:

[joshgoebel]

Thanks for that example. That looks wrong. I’m also not sure I’d call that a literal though. But it seems perhaps the exclamation rule should come before the number rule.

; Some unnamed metadata nodes, which are referenced by the named metadata.
!0 = !{!"zero"}
!1 = !{!"one"}
!2 = !{!"two"}
; A named metadata.
!name = !{!0, !1, !2}

[allejo]

Yea, I wouldn't call it a literal either. That usage is the closest thing I could find to it being considered something other than a number.

[joshgoebel]

Fixed.