Incompatibility with package-quickstart

[monnier]

if you create a ~/.emacs.d/package-quickstart.elfile withM-x package-quickstart-refresh`, you may see startup error messages like:

Error during quickstart: (file-missing "Cannot open load file" "Aucun fichier ou dossier de ce type" "<...>/.emacs.d/generic/proof-site")

This is because of the use of load-file-name within an eval-and-compile inside an autoloaded chunk of code.
The patch below seems to fix it (and it also takes advantage of macroexp-file-name when available).

diff --git a/proof-general.el b/proof-general.el
index c1f4c9127f..b948fe2172 100644
--- a/proof-general.el
+++ b/proof-general.el
@@ -3,7 +3,7 @@
 ;; This file is part of Proof General.
 
 ;; Portions © Copyright 1994-2012  David Aspinall and University of Edinburgh
-;; Portions © Copyright 2003-2022  Free Software Foundation, Inc.
+;; Portions © Copyright 2003-2024  Free Software Foundation, Inc.
 ;; Portions © Copyright 2001-2017  Pierre Courtieu
 ;; Portions © Copyright 2010, 2016  Erik Martin-Dorel
 ;; Portions © Copyright 2011-2013, 2016-2017  Hendrik Tews
@@ -55,19 +55,21 @@
 ;;   mostly mode hooks and autoloads).
 
 ;;;###autoload
-(eval-and-compile
-  (defvar pg-init--script-full-path
-    (or (bound-and-true-p byte-compile-current-file)
-        (and load-in-progress load-file-name)
+(defmacro pg-init--pg-root ()
+  `(file-name-directory
+    ,(if (fboundp 'macroexp-file-name) ;; Emacs≥28
+         `(macroexp-file-name)
+       `(or (bound-and-true-p byte-compile-current-file)
+            (and load-in-progress load-file-name)
+            buffer-file-name))))
 
-        (buffer-file-name)))
-  (defvar pg-init--pg-root
-    (file-name-directory pg-init--script-full-path)))
 
 ;;;###autoload
 (unless (bound-and-true-p byte-compile-current-file)
-  ;; This require breaks compilation, so it must only run when loading the package.
-  (require 'proof-site (expand-file-name "generic/proof-site" pg-init--pg-root)))
+  ;; This `require' breaks compilation, so it must run only when loading
+  ;; the package.
+  (require 'proof-site (expand-file-name "generic/proof-site"
+                                         (pg-init--pg-root))))
 
 (eval-when-compile
   ;; FIXME: [ This is used during installation of the ELPA package ]
@@ -81,7 +83,7 @@
            "coq" "easycrypt" "pghaskell" "pgocaml" "pgshell" "phox" "qrhl"
            )))
     (dolist (dir byte-compile-directories)
-      (add-to-list 'load-path (expand-file-name dir pg-init--pg-root)))))
+      (add-to-list 'load-path (expand-file-name dir (pg-init--pg-root))))))
 
 (provide 'proof-general)
 ;;; proof-general.el ends here

[monnier]

Nah, scratch that, it doesn't fix it either.

[monnier]

OK, this one seems to fare better (and I believe my test does now test what I intended):

diff --git a/proof-general.el b/proof-general.el
index c1f4c9127f..72945239e2 100644
--- a/proof-general.el
+++ b/proof-general.el
@@ -3,7 +3,7 @@
 ;; This file is part of Proof General.
 
 ;; Portions © Copyright 1994-2012  David Aspinall and University of Edinburgh
-;; Portions © Copyright 2003-2022  Free Software Foundation, Inc.
+;; Portions © Copyright 2003-2024  Free Software Foundation, Inc.
 ;; Portions © Copyright 2001-2017  Pierre Courtieu
 ;; Portions © Copyright 2010, 2016  Erik Martin-Dorel
 ;; Portions © Copyright 2011-2013, 2016-2017  Hendrik Tews
@@ -54,20 +54,15 @@
 ;; * Loading a minimal PG setup on startup (not all of Proof General, of course;
 ;;   mostly mode hooks and autoloads).
 
-;;;###autoload
-(eval-and-compile
-  (defvar pg-init--script-full-path
-    (or (bound-and-true-p byte-compile-current-file)
-        (and load-in-progress load-file-name)
-
-        (buffer-file-name)))
-  (defvar pg-init--pg-root
-    (file-name-directory pg-init--script-full-path)))
-
 ;;;###autoload
 (unless (bound-and-true-p byte-compile-current-file)
-  ;; This require breaks compilation, so it must only run when loading the package.
-  (require 'proof-site (expand-file-name "generic/proof-site" pg-init--pg-root)))
+  ;; This require breaks compilation, so it must run only when loading
+  ;; the package.
+  (require 'proof-site
+           (expand-file-name "generic/proof-site"
+                             (file-name-directory
+                              (or (and load-in-progress load-file-name)
+                                  buffer-file-name)))))
 
 (eval-when-compile
   ;; FIXME: [ This is used during installation of the ELPA package ]
@@ -76,12 +71,17 @@
   ;; session, so we here add to load-path all the subdirectories so
   ;; that files in (say) coq/ can (require 'coq-foo) and the compiler will find
   ;; the corresponding file.
-  (let ((byte-compile-directories
+  (let ((pg-root
+         (file-name-directory
+          (or (bound-and-true-p byte-compile-current-file)
+              (and load-in-progress load-file-name)
+              buffer-file-name)))
+        (byte-compile-directories
          '("generic" "lib"
            "coq" "easycrypt" "pghaskell" "pgocaml" "pgshell" "phox" "qrhl"
            )))
     (dolist (dir byte-compile-directories)
-      (add-to-list 'load-path (expand-file-name dir pg-init--pg-root)))))
+      (add-to-list 'load-path (expand-file-name dir pg-root)))))
 
 (provide 'proof-general)
 ;;; proof-general.el ends here

[erikmd]

Hi @monnier, thanks for your report and the patches!

The code you touched was initially written by @cpitclaudel so I Cc him in case he'd like to comment as well.

---

AFAIAC, I wasn't aware of package-quickstart but I see this is a new emacs feature (since 27.1) that is documented here.

I believe it'd be nice indeed to refactor this startup code provided this is backed up by CI tests.

So I have 2 questions (Cc @hendriktews, feel free to comment if you have an opinion!)

1. what test should we add to the CI to test that the refactor indeed fixes the issue?

2. is it needed to add some specific tests to ensure the refactor does not break existing "workflows"? (including MELPA's build (for instance, I've been told about this project: https://github.com/riscy/melpazoid (developed by one of the MELPA maintainers))), or not and I'm just worrying too much.

[monnier]

So I have 2 questions (Cc @hendriktews, feel free to comment if you have an opinion!) 1. what test should we add to the CI to test that the refactor indeed fixes the issue?

Good question. Currently, I can reproduce the problem in the following way: - M-x package-quickstart-refresh RET - rm ~/.emacs.d/package-quickstart.elc - Run `emacs`. Apparently when `~/.emacs.d/package-quickstart.el` gets compiled (which is done automatically in recent Emacsen), the problem does not appear.

2. Is it needed to add some specific tests to ensure the refactor does not break existing workflows?

Most likely, yes. Several years ago I had proposed and installed a fairly different approach to this part of the initialization and it broke some other workflow so it got reverted, but I don't think we recorded the particular setup in a test. 🙁