François Fleur

> The solution seems to be scanning the available snippets and including trigger characters for first characters that aren't alphanumeric (; in this case) Alternatively, you can define them manually,...

We can safely close this issue since PR #1813 got merged 5 days ago :)

It seems you have found the culprit. @greg-grigorjevs seems to use `windwp/nvim-ts-autotag` as well. I will close this issue (again), and happily reopen if someone provides a reproducer.

Hopefully this should covers most of the usual cases for buffer completion, such as substitutions and global commands. There are some commands (like `:vimgrep` or `:match`) that aren’t included here,...

I completely overlooked that the sources can be overridden :facepalm: I agree with your approach, it make more sense, I'll update the PR accordingly!

I force pushed an update, could you take another look? If I'm not mistaken, this should do what you meant.

Ah I hadn’t paid close attention to use strictly `buffer:enabled()` when you wrote it... but now I get it! > It seems like there's something wrong [...] as the menu...

> only substitution seems to be off by one char This is caused by `vim.on_key` ~~but I'm a bit clueless here.~~ For instance, while typing `:s/leo`, `vim.on_key` reports `s`, `/`,...

There is one drawback to your change in [`5427010`](https://github.com/Saghen/blink.cmp/pull/1734/commits/5427010386dd6dfd02b9a6fac181a57c23a5add7). If blink has not yet been initialized and the user triggers a keymap for substitute and init blink at the same...

> Would adding [...], alongside the existing vim.on_key, be enough? No, the issue is that `buffer.new` is called too late — specifically, after the first character is entered on the...