Adding support for regex as keys when expanding snippets

[Zetagon]

I read https://castel.dev/post/lecture-notes-1/ and wanted to replicate at least the simpler parts of the fraction snippet. Is it possible to easily write such snippets in yasnippet where the key is a regex?

In case it is not currently possible I have hacked together some elisp code using the interface provided that can do this but it is not very integrated into yasnippet. I have basically made a new keybinding that checks if a regex has matched. If it matched a corresponding snippet is expanded using yas-lookup-snippet and yas-expand-snippet. The configuration still happens in elisp code and not in a snippet buffer which is suboptimal.

Is anyone else interested in using regex as keys? If so I could with some help integrate this more into yasnippet and merge it.

[npostavs]

Is it possible to easily write such snippets in yasnippet where the key is a regex?

It's not currently possible.

Sounds like an interesting feature, though I think it breaks the current decision of looking up the snippet key in a hashtable, so implementation might be somewhat involved. I guess the check against regexps could just happen first, and then fallback to hashtable lookup otherwise.

#114 (Multiple keys) is sort of related.

[Zetagon]

My current approach is to do a regex match on from the beginning of the line to the cursor so that the search space is small. The regex is appended with a '$' so that the matched part is guaranteed to end at the cursor. The matched string is then deleted and placed in a global variable so that the snippet can use it. yas-expand-snippet is then used to find a snippet in the current mode.

I could submit it as a pull-request but I feel like it would be nicer if the regex could be defined in the snippet file instead of in elisp code.

[npostavs]

The regex is appended with a '$' so that the matched part is guaranteed to end at the cursor.

You can use \= instead. (elisp) Regexp Backslash

\=
matches the empty string, but only at point. (This construct is
not defined when matching against a string.)

---

it would be nicer if the regex could be defined in the snippet file instead of in elisp code.

Agreed.

[npostavs]

I could submit it as a pull-request but I feel like it would be nicer if the regex could be defined in the snippet file instead of in elisp code.

How would you feel about submitting a pull-request implementing the regex-in-snippet-file approach?

[Zetagon]

I would love to! If you could give me a few pointers to bits of code that are relevant I think I could do it.

[npostavs]

Okay, first off I should mention that since yasnippet is in GNU ELPA, you need to assign copyright before I can install subsantial (roughly defined as more than 15 lines) patches from you. To start the copyright assignment process, fill in this form, with "Emacs" as the program name, and send it to the mentioned email address.

---

As for code pointers, I'm thinking you'll want to allow the key field in the yas--template struct to be a list, with elements being a string (plain key) or (:regexp REGEXP) where REGEXP is the regexp key as a string.

yas-define-snippets and yas--add-template should be updated to handle this. For the latter, I think it should maintain a :regexp entry with an alist of (REGEXP . NAME) in the snippet key table (since we can't use hash lookup for regexps anyway). Then yas--fetch will have to iterate over the regexps and return the matching entries.

And finally, update yas--parse-template to accept a # regexp-key: header, or similar.

Those are my thoughts about it without trying to implement anything, so it's possible I've made some design bug in the above: if something doesn't make sense, don't hesitate to ask questions.

[netjune]

I read https://castel.dev/post/lecture-notes-1/

So many cool features of Ul­tiSnips. Hope they will be implemented in yasnippet.

[Zetagon]

So I have realized that the process for expanding regular snippets and regexp snippets will probably be different. I am thinking that regexp snippets should delete the text that matched the text and put that text into a variable like yas-matched-text that the snippet could use.

I think that it might be better to modify yas-expand or some other function that deals with expanding a snippet.

[npostavs]

I am thinking that regexp snippets should delete the text that matched the text

I think this part is already covered in yas-expand-snippet according to the BEG and END arguments. This is passed down from yas-expand-from-trigger-key, see the comment ;; Delete snippet key and active region when expanding.

put that text into a variable like yas-matched-text that the snippet could use.

This part is missing though.

[Zetagon]

I noticed that yas-expand-from-trigger-key expects one position and multiple templates which is not compatible with regexps because the text matched by the regexps don't have to be of the same length. The easiest solution seems to be to make a list of positions for each template.

[npostavs]

The easiest solution seems to be to make a list of positions for each template.

It doesn't look like any callers are interested in any template apart from the first, maybe it's easier to always return a single template, with a single position.

[Zetagon]

Sorry if this is off topic but how do I debug tests in yasnippet? If I run ert as is described in the test file I just get the message that there are unexpected errors. I tested running ert manually so that I can get a backtrace but the backtrace is so large that I can't even copy the backtrace without Emacs hanging or lagging immensly.

[npostavs]

If I run ert as is described in the test file

Oh, maybe it's because ert-run-tests-batch-and-exit doesn't exist in the old ert that works with Emacs 22 (I'm not actually sure if yasnippet works in such an old Emacs), but normally it should look like this:
emacs --batch -Q -L . -l yasnippet-tests.el -f ert-run-tests-batch-and-exit

the backtrace is so large that I can't even copy the backtrace without Emacs hanging or lagging immensly.

Which Emacs version are you using? I think there was a bug in 26.1 that caused it to ignore print-length when making backtraces (should be fixed in 26.2, if I recall correctly). You can probably work around this by setting debugger-print-function to prin1 (as it was in earlier Emacs versions).

Alternatively, use emacs -Q -L . -l yasnippet-tests.el -f ert to test interactively. Then you can use . on the failing tests to jump their definitions, instrument them with edebug and step through (note that the snippet loading tests pass only once in an Emacs session because they rely on global state, it's something I've been meaning to fix, but never reaches high enough priority).

---

If you have rake installed, you can use rake tests and rake itests to run those command lines.