dylan-lang / dylan-emacs-support Goto Github PK

M-x dime
open strings
build
,change-package strings
C-x C-f strings.dylan
C-o

Evaluation aborted on Error:
swank:buffer-first-change is not present as a key for
#[(swank:set-package -> {method ( ())}),
(swank:dylan-subclasses -> {method ( ())}), (swank:connection-info -> {method ()}),
(swank:describe-symbol -> {method ( ())}), (swank:default-directory -> {method ()}),
(swank:dylan-superclasses -> {method ( ())}),
(swank:create-repl -> {method ( ())}),
(swank:find-definitions-for-emacs -> {method ( ())}),
(swank:set-default-directory -> {method ( ())}), (swank:quit-lisp -> {method ()}),
(swank:listener-eval -> {method ( ())}),
(swank:compile-file-for-emacs -> {method ( ())}),
(swank:list-all-package-names -> {method ( ())}),
(swank:xref -> {method ( (, ))}),
(swank:operator-arglist -> {method ( (, ))}),
(swank:compiler-notes-for-emacs -> {method ()})].
.

in the expression let foo-bar = 24; foo gets highlighted, but -bar doesn't. It works fine if you write let foo-bar :: <integer> = 24. A similar issue appears in slots:

define class <foo> (<object>)
  slot foo-bar
end;

vs

define class <foo> (<object>)
  slot foo-bar :: <integer>
end;

What we have now:

let foo = really-long-function-name(
                                    long-argument-1(),
                                    long-argument-2())

What I want:

let foo = really-long-function-name(
             long-argument-1(),
             long-argument-2())

Note, never this:

let foo = really-long-function-name(argument-0-on-same-line(),
             long-argument-1(),
             long-argument-2())

If there's an argument immediately following the open paren then subsequent args should line up under the first (as they do currently):

let foo = really-long-function-name(long-argument-1(),
                                    long-argument-2())

The OD DIME docs say that parameter lists should be displayed when you type ( and when you type , inside a parameter list. This isn't happening for me despite that M-. works, so I know the connection is good and the project compiled correctly.

I didn't see any errors, nor any messages that looked relevant in *inferior-dylan*.

• version: 9d2891e
• macOS Sonoma 14.4
• emacs 29.2

The body of define test should all indent 2 spaces. Instead, the first body line indents 4 spaces.

define test test-foo ()
    assert-equal(...);
  assert-equal(...);

This is probably my biggest dylan-mode annoyance.

Using DIME @ version 9d2891e these warnings appear in the *Messages* buffer:

Loading /Users/cgay/dylan/workspaces/dylan-emacs-support/dylan.el (source)...done
dime.el: Warning: Use keywords rather than deprecated positional arguments to `define-minor-mode' [5 times]
dime.el: Warning: Obsolete calling convention for 'sit-for'
dime-repl.el: Warning: Use keywords rather than deprecated positional arguments to `define-minor-mode' [2 times]

In the recent commit 127bead, destructure-case was renamed to dime--destructuring-case everywhere except for its definition where it was called dime-destructuring-case. The definition should be changed to match the uses (that symbol should not be accessed externally, so the double-dash is correct)
definition:
https://github.com/dylan-lang/dylan-mode/blob/127bead9241060bbfb0a0e5dac6e7cfe32708919/dime.el#L614
example of use:
https://github.com/dylan-lang/dylan-mode/blob/127bead9241060bbfb0a0e5dac6e7cfe32708919/dime.el#L1991

Files with Dime prefix

• dime-browse.el -- good authorship info, GPL version 2 or later.
• dime-note-tree.el -- who wrote this? GPL version 2 or later.
• dime-repl.el -- file says main author is Helmut Eller, GPL version 2 or later.
• dime.el -- file lists several authors, GPL version 2 or later.

Files with Dylan prefix

• dylan-lid.el -- file says it's written by Erik Charlebois, GPL version 1 or later.
• dylan-mode.el -- good authorship info, GPL version 1 or later.
• dylan-optimization-coloring.el -- no license info at all, git log shows it's mainly Hannes Mehnert's work.

When this happens a given file will at first not be highlighted.

After killing the affected buffer and re-finding the file, hightlighting works.

dylan-mode highlights macros like with-open-file. If you import file-system with prefix "fs/" (or probably anything else) the highlighting of fs/with-open-file doesn't happen. To solve this I suppose we need a standard module prefix separator. I actually don't like "/" as in "fs/", but it's the one that sucks less than the other available options, in my opinion.

Currently we have two sets of files: dylan-*.el and dime-*.el. There seems to be a somewhat clean separation between Dime and the other Dylan stuff. The Dime stuff depends on the Dylan stuff, but the Dylan stuff can also be used standalone.

In light of this division, I think we should split into two different Melpa packages: dylan and dime, with dylan as a dependency for dime. Both package names are available so there is no conflict with existing packages.

We can still keep both packages in the same git repo. A Melpa recipe can be written to only pick files matching a shell glob from a repo.

So I set up DIME with this code

;;; Dylan
(add-to-list 'load-path "~/dylan-mode")
(setq inferior-dylan-program "dswank")
(require 'dime)
(dime-setup '(dime-dylan dime-repl))
(setenv "OPEN_DYLAN_USER_REGISTRIES" "/home/rage/Downloads/opendylan-2014.1/sources/registry")

but when I try to run DIME it claims that I never set inferior-dylan-program. Am I doing something wrong or is the plugin broken somehow.

The dtags script relies on the gema macro processor which is not easy to install on any major OS.

If people are still using dtags, it would probably be a good idea to rewrite it using either Dylan or Emacs Lisp. At a glance, it looks like a job that Elisp would do well at.

Ubuntu 12.10 ships with Emacs 24.1.1, and DIME/dylan-mode does not work. The issues I'm looking at right now include:

• defvar-local does not exist
• cl-find-if does not (appear to) exist

I'm fixing these in my fork and will make a pull request once things are copacetic.

In the recent commit 127bead, destructure-case was renamed to dime--destructuring-case everywhere except for its definition where it was called dime-destructuring-case. The definition should be changed to match the uses (that symbol should not be accessed externally, so the double-dash is correct)
definition:
https://github.com/dylan-lang/dylan-mode/blob/127bead9241060bbfb0a0e5dac6e7cfe32708919/dime.el#L614
example of use:
https://github.com/dylan-lang/dylan-mode/blob/127bead9241060bbfb0a0e5dac6e7cfe32708919/dime.el#L1991

This package was an early submission to MELPA and doesn't follow modern packaging conventions. Do you accept PRs to fix that?

The main issue is namespace hygiene. Most urgently, dime.el defines its own versions of common macros without using a dime- namespace prefix for them. Of those, at least when-let is incompatible with the standard definition of that macro and breaks other Emacs packages.

DIME also defines some things using a sldb- prefix; that prefix belongs to the SLIME debugger and DIME should use something else.

Would you be willing to raise the minimum required Emacs version to 25.1 which is four years old? That would get us common macros like when-let via (require 'subr-x).

Other issues:

• Lexical binding should be used throughout.
• Package metadata should be added as headers into the .el files.
• Files and functions/macros/variables should be renamed such that the definition prefixes correspond to the filenames in which they are defined.

I have several packages in MELPA and am happy to assist with all of this stuff if welcome. To not break other packages, at least the namespace hygiene should be fixed.

Sometimes, when I type a semicolon, dylan-mode re-indents multiple lines of code. Now, if it were able to always get the indentation exactly right I might consider this correct behavior, but it doesn't and I don't. I personally would prefer not to have it do any indentation except when I press the Tab key. (Auto-indent after pressing Enter would be okay too, but not my preference.)

Here's a specific and particularly egregious instance of the problem. Given the following chunk of code...

define method parse-file
    (parser :: <coil-parser>, struct :: <struct-prototype>) => (struct :: <struct-prototype>)
  local method fail ()
          parse-error(parser, "Target of @file keyword must be a filename or a "
                        "list of the form [ 'filename' 'path.to.struct' ].");
        end;
  let target = parse-any(parser);
  let path = "@root";
  select (target by instance?)
    <string> => #f;
    <sequence> =>
      if (target.size ~= 2
            | ~every?(rcurry(instance?, <string>), target))
        fail();
      else
        path := target[1];
        target := target[0];
      end
    otherwise => fail();
  end;

if I put the cursor at the end of the otherwise => fail(); line, erase the semicolon and type it again, the code ends up like this:

define method parse-file
    (parser :: <coil-parser>, struct :: <struct-prototype>) => (struct :: <struct-prototype>)
  local method fail ()
          parse-error(parser, "Target of @file keyword must be a filename or a "
                        "list of the form [ 'filename' 'path.to.struct' ].");
        end;
  let target = parse-any(parser);
  let path = "@root";
  select (target by instance?)
    <string> => #f;
    <sequence> =>
    if (target.size ~= 2
          | ~every?(rcurry(instance?, <string>), target))
      fail();
    else
      path := target[1];
      target := target[0];
    end
      otherwise => fail();
  end;

Note that code all the way up to the "if" has been re-indented and it has been moved flush with the <sequence> => line. The only way I could get a semicolon there was to cut and paste it since that didn't invoke the indent behavior.

Leaving this here as a reminder. Could be useful but I haven't tried it.
http://www.flycheck.org/en/latest/index.html

Compiling dime-20210319.1414 on Emacs 27.1 I get these warnings:

In dime-repl-mode:
dime-repl.el:414:9:Warning: assignment to free variable
    ‘dime-write-string-function’
dime-repl.el:1488:7:Warning: assignment to free variable
    ‘dime-output-target-to-marker-function’

In end of data:
dime-repl.el:1496:1:Warning: the function ‘dime-emit-to-target’ is not known
    to be defined.

Not sure if this impacts the functionality, I haven't found anything yet.

Ref #41

DIME was forked from SLIME, and DIME currently defines symbols using SLIME's sldb- prefix. We should change the prefix.

In fact, SLIME itself is not following current Emacs Lisp namespace conventions since it uses sldb instead of slime as the prefix. Since we can start afresh with DIME we should follow the convetions.

I suggest changing sldb to dime-debug. The following would be the list of the new symbol names. Does it look reasonable?

• dime-debug-insert-frame-call-to-repl
• dime-debug-hook
• dime-debug-initial-restart-limit
• dime-debug-condition
• dime-debug-restarts
• dime-debug-level
• dime-debug-backtrace-start-marker
• dime-debug-restart-list-start-marker
• dime-debug-continuations
• dime-debug-mode-syntax-table
• dime-debug-mode
• dime-debug-buffers
• dime-debug-find-buffer
• dime-debug-get-default-buffer
• dime-debug-get-buffer
• dime-debug-debugged-continuations
• dime-debug-setup
• dime-debug-activate
• dime-debug-reinitialize
• dime-debug-exit
• dime-debug-close-step-buffer
• dime-debug-insert-condition
• dime-debug-extras-hooks
• dime-debug-dispatch-extras
• dime-debug-insert-restarts
• dime-debug-insert-more-restarts
• dime-debug-frame-string
• dime-debug-frame-number
• dime-debug-frame-plist
• dime-debug-frame-restartable-p
• dime-debug-prune-initial-frames
• dime-debug-insert-frames
• dime-debug-compute-frame-face
• dime-debug-insert-frame
• dime-debug-fetch-more-frames
• dime-debug-restart-at-point
• dime-debug-frame-number-at-point
• dime-debug-var-number-at-point
• dime-debug-previous-frame-number
• dime-debug-frame-details-visible-p
• dime-debug-frame-region
• dime-debug-forward-frame
• dime-debug-backward-frame
• dime-debug-goto-last-frame
• dime-debug-beginning-of-backtrace
• dime-debug-recenter-region
• dime-debug-default-action
• dime-debug-default-action/mouse
• dime-debug-cycle
• dime-debug-end-of-backtrace
• dime-debug-fetch-all-frames
• dime-debug-show-source
• dime-debug-show-frame-source
• dime-debug-toggle-details
• dime-debug-show-frame-details
• dime-debug-frame-details
• dime-debug-insert-frame-variable-value-function
• dime-debug-insert-locals
• dime-debug-insert-frame-variable-value
• dime-debug-hide-frame-details
• dime-debug-disassemble
• dime-debug-eval-in-frame
• dime-debug-pprint-eval-in-frame
• dime-debug-inspect-in-frame
• dime-debug-inspect-var
• dime-debug-inspect-condition
• dime-debug-down
• dime-debug-up
• dime-debug-sugar-move
• dime-debug-details-up
• dime-debug-details-down
• dime-debug-quit
• dime-debug-continue
• dime-debug-abort
• dime-debug-invoke-restart
• dime-debug-invoke-restart-by-name
• dime-debug-break-with-default-debugger
• dime-debug-break-with-system-debugger
• dime-debug-step
• dime-debug-next
• dime-debug-out
• dime-debug-break-on-return
• dime-debug-break
• dime-debug-return-from-frame
• dime-debug-restart-frame
• dime-debug-recompile-frame-source

The second line doesn't indent correctly (or at all):

  if (starts-with?(url, "https://github.com"))
make(<git-transport>)

If I change "https://github.com" to the empty string dylan-mode indents the second line correctly.

Bugs / to-do list

• Don't highlight macro variables (e.g., "?x:" in ?x:name) as keywords.
  Just highlight the "x" as a variable binding.

• Customize fill-column to some acceptable value so that auto-filled
  comments are filled at a standard place > 70. I use 89 myself.

• It appears as though some code matches only some of the graphic-character
  BNF when it should match all graphic chars.

The majority of sources/lib/big-integers/double-integer.dylan appears to be commented out because it uses dylan/* as the name for * imported from the dylan module.

(This is obviously low priority.)

Quoting @housel from #41 (comment)

The dylan-optimization-coloring.el code isn't directly integrated; it can be used on its own to read files generated using the compiler's -dispatch-coloring elisp option.

and #41 (comment)

With optimization-coloring-mode, two files are involved. The main file is the Dylan source file, which starts out using dylan-mode. If you compile using the -dispatch-coloring elisp option, for each source file you get an output file (using elisp syntax) that contains annotations for dispatch coloring (described, in the IDE case, in https://opendylan.org/documentation/getting-started-ide/coloring.html#about-dispatch-optimizations). The dylan-optimization-coloring.el code can be run on an existing dylan-mode buffer to replace the major mode's coloring with coloring based on the output from the compiler.

So yes, there should be a dylan-optimization-coloring-mode, it's just that nobody has ever spent the time to restructure the code as a minor mode.

File compilation through dime fails with an error,

Evaluation aborted on {<type-error> : #f is not of type {<class>: <project-object>}}.

I've tried importing, opening and building all through the repl and that is working perfectly, but I can't compile a file from the buffer with C-c C-k without hitting this error.

I've created a Vagrant environment (Ubuntu, Emacs 24.3.1) where this can be reproduced, https://github.com/danmidwood/dime-error-repro

Adding any of these adjectives, other than "inline" causes syntax coloring to fail. (It stops coloring the word "define".)

    #"default-inline" => 0;
    #"inline"         => 1;
    #"inline-only"    => 2;
    #"may-inline"     => 3;
    #"not-inline"     => 4;

SLIME transmits always the current package over the wire -- so does DIME.

But Dylan has the concept of both libraries (unit of compilation), modules (namespace); and additionally Open Dylan imposes an "active project" (the one we're interested in).

These 3 things make sense to a certain extend: names are mangled regarding the project, library and module. Unfortunately both the emacs and dylan side only transmit a single piece of information - the module. This should be modified so that the full information is transmitted for each request. This would minimize state on one of the sides, and therefore less stuff can get out of sync.

define compiler-sideways method print-method
    (stream :: <stream>, o :: <&method>, #key outputter)
end;

the body currently wants to be aligned to method + 2, whereas my intuition is it should be indented by 2 only...

try module: foo2 and you'll get red highlighted text