bug#66845: 29.1; cmake-ts-mode indentation broken

[Randy Taylor <dev@rjt.dev>]

[-- Attachment #1: Type: text/plain, Size: 2627 bytes --]

On Tuesday, October 31st, 2023 at 07:35, Juan Palacios <jpalaciosdev@gmail.com> wrote:
> Using cmake grammar from https://github.com/uyha/tree-sitter-cmake
> 
> Reproduction:
> 1. emacs -q
> 2. C-x C-f test.cmake
> 3. M-x cmake-ts-mode
> 4. Insert the following text:
> if(TRUE)
> message("indent")
> endif()
> 5. Move the point to the beginning of the second line.
> 6. TAB
> 7. Notice that no indentation is performed.
> 
> treesit-simple-indent-rules local value:
> ((cmake
> ((node-is ")")
> parent-bol 0)
> ((node-is "else_command")
> parent-bol 0)
> ((node-is "elseif_command")
> parent-bol 0)
> ((node-is "endforeach_command")
> parent-bol 0)
> ((node-is "endfunction_command")
> parent-bol 0)
> ((node-is "endif_command")
> parent-bol 0)
> ((parent-is "foreach_loop")
> parent-bol cmake-ts-mode-indent-offset)
> ((parent-is "function_def")
> parent-bol cmake-ts-mode-indent-offset)
> ((parent-is "if_condition")
> parent-bol cmake-ts-mode-indent-offset)
> ((parent-is "normal_command")
> parent-bol cmake-ts-mode-indent-offset)))
> 
> treesit-explore-mode shows:
> (if_condition
> (if_command (if) (
> (argument_list
> (argument (unquoted_argument)))
> ))
> (body
> (normal_command (identifier) (
> (argument_list
> (argument
> (quoted_argument " (quoted_element) ")))
> )))
> (endif_command (endif) ( )))
> 

I've attached a patch fixing this while retaining compatibility with
older grammars. 2 new versions of the grammar, v0.3.0 and v0.4.0
(and anything beyond for now) wrap things in new nodes.

Yuan,
It's been awhile since I've touched any tree-sitter stuff so I'm
probably missing something obvious, but it seems to me like the
function I've added in the patch
(cmake-ts-mode--argument-list-body-offset) is working around a bug
somewhere? I'm not sure why the start of line_comment is considered
parent-bol for indentation.

For example, with this test cmake file: (test.cmake)
if(TRUE)
  # Comment.
  message("indent")
endif()

We get:
(source_file
 (if_condition
  (if_command (if) (
   (argument_list
    (argument (unquoted_argument)))
   ))
  (body (line_comment)
   (normal_command (identifier) (
    (argument_list
     (argument
      (quoted_argument " (quoted_element) ")))
    )))
  (endif_command (endif) ( ))))

With cmake-ts-mode--argument-list-body-offset always returning
cmake-ts-mode-indent-offset (which is what I would expect), the
`parent-is "body"` rule will cause this indentation:
if(TRUE)
  # Comment.
    message("indent")
endif()

Am I being silly or is there indeed something odd going on?

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: 0001-Fix-cmake-ts-mode-indentation-Bug-66845.patch --]
[-- Type: text/x-patch; name=0001-Fix-cmake-ts-mode-indentation-Bug-66845.patch, Size: 3053 bytes --]

From 544396b7afa55c4d12fc6f03198199017582b8df Mon Sep 17 00:00:00 2001
From: Randy Taylor <dev@rjt.dev>
Date: Tue, 31 Oct 2023 22:08:25 -0400
Subject: [PATCH] Fix cmake-ts-mode indentation (Bug#66845)

* lisp/progmodes/cmake-ts-mode.el
(cmake-ts-mode--argument-list-body-offset): New function.
(cmake-ts-mode--indent-rules): Support versions v0.3.0 and v0.4.0 of
the grammar.
(cmake-ts-mode--font-lock-compatibility-fe9b5e0): Fix docstring.
---
 lisp/progmodes/cmake-ts-mode.el | 30 ++++++++++++++++++++++++++++--
 1 file changed, 28 insertions(+), 2 deletions(-)

diff --git a/lisp/progmodes/cmake-ts-mode.el b/lisp/progmodes/cmake-ts-mode.el
index 53d471c381a..d2d66a461f3 100644
--- a/lisp/progmodes/cmake-ts-mode.el
+++ b/lisp/progmodes/cmake-ts-mode.el
@@ -52,6 +52,24 @@ cmake-ts-mode--syntax-table
     table)
   "Syntax table for `cmake-ts-mode'.")
 
+(defun cmake-ts-mode--argument-list-body-offset (_n parent &rest _)
+  "This offset is used for argument_list and body nodes.
+
+For some reason, a comment after an argument_list or body node causes
+the next line to indent as if the comment was the parent. For example:
+
+1 if (TRUE)
+2   # Comment.
+3     func()
+4 endif()
+
+Line 3 is indented as if the comment (line 2) was parent-bol,
+when line 1 should be parent-bol."
+  (when-let (node (treesit-node-child parent 0 t))
+    (if (string-equal "line_comment" (treesit-node-type node))
+        0
+      cmake-ts-mode-indent-offset)))
+
 (defvar cmake-ts-mode--indent-rules
   `((cmake
      ((node-is ")") parent-bol 0)
@@ -60,6 +78,14 @@ cmake-ts-mode--indent-rules
      ((node-is "endforeach_command") parent-bol 0)
      ((node-is "endfunction_command") parent-bol 0)
      ((node-is "endif_command") parent-bol 0)
+     ;;; Release v0.4.0 wraps arguments in an argument_list node.
+     ,@(ignore-errors
+         (treesit-query-capture 'cmake '((argument_list) @capture))
+         `(((parent-is "argument_list") parent-bol cmake-ts-mode--argument-list-body-offset)))
+     ;;; Release v0.3.0 wraps the body of commands into a body node.
+     ,@(ignore-errors
+         (treesit-query-capture 'cmake '((body) @capture))
+         `(((parent-is "body") parent-bol cmake-ts-mode--argument-list-body-offset)))
      ((parent-is "foreach_loop") parent-bol cmake-ts-mode-indent-offset)
      ((parent-is "function_def") parent-bol cmake-ts-mode-indent-offset)
      ((parent-is "if_condition") parent-bol cmake-ts-mode-indent-offset)
@@ -89,8 +115,8 @@ cmake-ts-mode--if-conditions
   "CMake if conditions for tree-sitter font-locking.")
 
 (defun cmake-ts-mode--font-lock-compatibility-fe9b5e0 ()
-  "Indent rules helper, to handle different releases of tree-sitter-cmake.
-Check if a node type is available, then return the right indent rules."
+  "Font lock helper, to handle different releases of tree-sitter-cmake.
+Check if a node type is available, then return the right font lock rules."
   ;; handle commit fe9b5e0
   (condition-case nil
       (progn (treesit-query-capture 'cmake '((argument_list) @capture))
-- 
2.42.0