Tests failing under Emacs 29+ about yasnippet

[aagon]

A few precisions : the place where the snippet argument becomes nil is during the call to :

(delete-region (or from (yas--field-start field)) (yas--field-end field))

in the function yas--skip-and-clear.

In order to establish that, one would just need to call yas-active-snippets just before and after the aforementioned line.

Under emacs 29, this evaluates to non-nil before, and nil after, whereas under emacs 28, this evaluated to something non-nil both before and after.

[manphiz]

After instrumenting the function yas-active-snippets to observe the test process, it looks like under Emacs 28, after a series of calls to yas-active-snippets, the test will end after passing both beg and end with nil. However under Emacs 29, yas--on-protection-overlay-modification got triggered after that, printing a message of [yas] Committing snippets. Action would destroy a protection overlay., and yas-active-snippets gets invoked again with beg as 1 and end as 1, which results in snippet becomes nil. Hopefully someone familiar with yasnippet internal will figure out why yas--on-protection-overlay-modification got triggered here.

As there exists a customization of yas-inhibit-overlay-modification-protection, setting it to t helps letting both tests pass.

[dgutov]

If there are overlay-related behaviors that have changed between 28 and 29, emacs-devel will really want to know the details.

[Cons-Cat]

This appears to cause only the very last snippet key in .yas-compiled-snippets.el to work correctly, regardless of how many snippets are defined or in what order.

[monnier]

This appears to cause only the very last snippet key in `.yas-compiled-snippets.el` to work correctly, regardless of how many snippets are defined or in what order.

Does the problem show up in Emacs-30? It sounds like the issue we fixed in https://debbugs.gnu.org/65929 IIRC "we" were supposed to add a hack to YASnippet to work around the problem in Emacs-29.1.

[Cons-Cat]

Does the problem show up in Emacs-30?

Yes, I'm using

GNU Emacs 30.0.50
Development version 790b5982175b on master branch; build date 2024-01-09.

I could be entirely wrong that the bug I see is related to this issue, though. I thought so because I get a similar error message.

[monnier]

Hmm... that's odd. I just tried to run the tests, bumped into a bunch of problems which I now fixed on the main branch (of yasnippet), but all the tests that pass in Emacs-28.2 also pass in Emacs-30 for me (more specifically, for me all tests pass except `test-yas-in-org`, `test-yas-tab-binding`, and `issue-492-and-494`; this last one only fails in Emacs-28 but not in Emacs-30). I also tried using an Emacs built from the `emacs-29` branch, and got the same result. I'll try Emacs-29.1 when I'm on a machine that has this version handy.

[Cons-Cat]

As a sanity check, I've just run it under emacs -Q. I did package-install yasnippet, set my yas-snippet-dirs, and got the same issue, so it definitely doesn't relate to my other configurations. However, I don't seem to get the Wrong type argument: characterp, nil error message that I do in my real configuration, which seems super weird. That error message might have been completely been unrelated to this issue.

[monnier]

As a sanity check, I've just run it under `emacs -Q`. I did `package-install yasnippet`, set my `yas-snippet-dirs`, and got the same issue, so it definitely doesn't relate to my other configurations.

IIUC this was also with Emacs-30. Could you clarify what is "the same issue" exactly, or more specifically, what steps you took after setting `yas-snippet-dirs`, what happened and what you expected to happen instead? [ Just for clarification: I'm playing the role of maintainer, but I've never actually used YASnippet, so think of me as a kind of fresh new beginner user of YASnippet, even though I'm fairly knowledgeable about some aspects relevant to the maintenance 🙂 ]

[Cons-Cat]

Could you clarify what is "the same issue" exactly, or more specifically, what steps you took after setting yas-snippet-dirs, what happened and what you expected to happen instead?

Yes, I'm using Emacs 30 for this. Setting yas-snippet-dirs isn't really important here (it doesn't seem to affect this problem), only calling yas-define-snippets one way or another is important. I've designed a minimal reproduction case here:

I run simply:
emacs -Q

M-x package-delete yasnippet

M-x package-install yasnippet

Then in the scratch buffer, I evaluate:

(yas-minor-mode)
(yas-define-snippets 'emacs-lisp-mode
                     '(("a" "aaa" "function" nil
                        nil nil
                        "" nil
                        nil)
                       ("b"
                        "bbb"
                        "function" nil nil nil
                        "" nil
                        nil)))

I then type b, and press TAB to expand, which transforms it into bbb. If I type a and expand the same way, it does not appear to do anything. I would expect that both a and b expand correctly here.

[monnier]

I pushed to `master` a patch which should fix the two test failures under Emacs-29.1 (which I could indeed reproduce in Emacs-29.1).

``` (yas-define-snippets 'emacs-lisp-mode '(("a" "aaa" "function" nil nil nil "" nil nil) ("b" "bbb" "function" nil nil nil "" nil nil))) ``` I then type `b`, and press `TAB` to expand, which transforms it into `bbb`. If I type `a` and expand the same way, it does not work. I would expect that both a and b expand correctly here.

No big surprise here, tho: you gave the same name ("function") to the two snippets, so the second overrides the first. `yasnippet.el` should admittedly detect this and complain about the name clash, but other than that I think it behaves "as expected" here.

[Cons-Cat]

Aaaaaaa I feel so silly now! That was the issue in my real Emacs config as well. Thank you for pointing that out!

[monnier]

I still don't see why you wrote about your problems in this issue, since they seem wholly unrelated.
In any case, I think this issue is now closed.