else/elif seem unsupported?

[hseg]

Consider the following MWE:

cabal-version: 2.4
if os(mingw32)
  constraints: tar +static
else if os(mingw32)
  constraints: tar -static
elif os(mingw32)
  constraints: tar +static +win32

It is currently formatted as

cabal-version: 2.4

if os(windows)
  constraints: tar +static
else if os ( mingw32 )
  constraints: tar -static

elif os ( mingw32 )
  constraints: tar +static +win32

where I would expect it to be formatted as

cabal-version: 2.4

if os(windows)
  constraints: tar +static
elif os(windows)
  constraints: tar -static
elif os(windows)
  constraints: tar +static +win32

(note the spacing!)

[tfausak]

Thanks for reporting this issue! I think there's some confusion about how conditionals work in package descriptions. Prior to Cabal 2.2, elif wasn't supported at all. There was only if and else. So your input would be formatted like this:

if flag(a)
  constraints: tar +static
else
  if flag(b)
    constraints: tar -static
  else
    if flag(c)
      constraints: tar +static +win32

Note that nothing else can go on the line with else.

Starting with Cabal 2.2, the new elif section was allowed. So you could write your input like this:

cabal-version: 2.2

if flag(a)
  constraints: tar +static
elif flag(b)
  constraints: tar -static
elif flag(c)
  constraints: tar +static +win32

[hseg]

On Thu, Sep 12, 2024 at 06:40:39AM GMT, Taylor Fausak wrote: Note that nothing else can go on the line with `else`.

Ah, right! ```cabal if os(mingw32) constraints: tar +static else if os(mingw32) constraints: tar -static ``` indeed correctly gives ```cabal if os(windows) constraints: tar +static else if os(windows) constraints: tar -static ``` So `cabal-gild` is correctly not accepting an `else` section with anything on its line (though it'd be nice if it either warned about this or would automatically fix it).

Starting with Cabal 2.2, the new `elif` section was allowed.

Fine. These are the possibilities I see for this: - cabal-gild pretty-prints the `elif`, but warns if `cabal-version` is not specified to be at least 2.2 - cabal-gild automatically bumps the `cabal-version` to be at least 2.2 when an `elif` section is encountered - cabal-gild desugars the `elif` if the in-file `cabal-version < 2.2` and pretty-prints it otherwise - the same, but controlled by a command-line argument of what cabal version to target Given the project goals, and given we probably want to avoid being too clever, the last two options don't look very promising. Either of the first two is good by me.

[tfausak]

Gild is very intentionally only a formatter. It does not aim to provide any validation or linting. else foo is syntactically valid, so Gild formats it. Similarly, elif is syntactically valid even for older Cabal versions, so Gild formats it.

Cabal itself will emit an error if you provide any arguments to else. For example:

cabal-version: 2.0
name: example
version: 0
build-type: Simple

library
  if False
    x-whatever: 0
  else if False
    x-whatever: 1

Errors encountered when parsing cabal file ./example.cabal:

example.cabal:9:3: error:
`else` section has section arguments [SecArgName (Position 9 8) "if",SecArgName (Position 9 11) "False"]

    8 |     x-whatever: 0
    9 |   else if False
      |   ^

And it will warn about using elif with an unsupported version of Cabal:

cabal-version: 2.0
name: example
version: 0
build-type: Simple

library
  if False
    x-whatever: 0
  elif False
    x-whatever: 1

...
Warning: example.cabal:9:3: invalid subsection "elif". You should set
cabal-version: 2.2 or larger to use elif-conditionals.
...

[hseg]

On Thu, Sep 12, 2024 at 09:46:23AM GMT, Taylor Fausak wrote: Gild is very intentionally only a formatter. It does not aim to provide any validation or linting. `else foo` is syntactically valid, so Gild formats it. Similarly, `elif` is syntactically valid even for older Cabal versions, so Gild formats it.

OK. However, the formatting for `elif` as a directive starting a conditional section, versus a generic section-starting directive, *is* something where Gild could err on the side of newer cabal formats on, correct? My other musings on emitting the correct cabal-version aside, we could pretend `elif` is a supported directive, and have cabal catch if the usage was incorrect. (This would remove `else if -> elif` from the table, which is a compromise I'm willing to make)

[tfausak]

Are you asking for Gild to always format elif, even if the Cabal version is less than 2.2? It certainly could do that, but it deliberately doesn't: #83 (comment)

Once the user notices the warning from Cabal, presumably they'll edit their package description to use a recent-enough Cabal version. Then all they have to do is re-run Gild and it should be formatted correctly.

It's also worth noting that elif os(mingw32) and elif os ( mingw32 ) are semantically the same. The extra spaces don't change the meaning at all.

[hseg]

I'm saying that elif is never formatted, even if Cabal is at least 2.2, which it seems you believe is not what's happening? How is Gild detecting the cabal version, if so?
I was suggesting Gild always format elif for simplicity's sake, and to leave the correctness concern to Cabal.
And yes, elif os ( mingw32 ) is semantically identical, but I want my tools to improve the formatting, not worsen it.

[tfausak]

But elif does get formatted?

-- input
cabal-version: 2.2
elif flag ( BLAH )
  x-blah: true

-- output
cabal-version: 2.2
elif flag(blah)
  x-blah: true

There's also a test for this case:

cabal-gild/source/test-suite/Main.hs

Lines 660 to 663 in f30f35e

	Hspec.it "formats new elif" $ do
	expectGilded
	"cabal-version: 2.2\nelif flag ( x )"
	"cabal-version: 2.2\nelif flag(x)\n"

[hseg]

I could've sworn I could reproduce the issue last week, but looking back at my scrollback and the initial file, it seems I wasn't noticing the absence of a `cabal-version` header. My bad, sorry for the noise!

[tfausak]

No problem! Sorry for the confusion.