Fix emacs state corruption from incorrect "syntax-begin-function" setting

[MicahChalmer]

This is to fix #36.

Commit 6b57bbf set syntax-begin-function to beginning-of-defun. This violated the contract for that function--it's supposed to get outside of all braces/parens--but functions could appear inside other blocks, so beginning-of-defun could end inside another pair of braces. This in turn caused emacs corrupt some unknown internal state and cause itself to think there were fewer parens than there were.

The comments of the added test show how much I understood about the exact mechanism of the failure--which is not much. I figure it's better to just fix it and leave the test there. Though I don't understand the mechanism of the corruption, at least I understand how the code was violating the hook's contract. Hopefully there isn't some other similar corruption hiding out that I haven't found yet.

@vhbit and @hugoduncan, does this fix the issues you were seeing?

[rust-highfive]

r? @huonw

(rust_highfive has picked a reviewer for you, use r? to override)

[pnkfelix]

typo: "corruped"

[pnkfelix]

I have nits, but the fix seems as sound as far I understand it. So I'm going to land this PR, and file an issue asking @MicahChalmer to fix the comments. :)

[pnkfelix]

Oh, one more question: @MicahChalmer can you elaborate (or point to an elaboration elsewhere) on why #30 included that setq of syntax-begin-function to beginning-of-defun ? I.e. do you remember if there was a specific bug that you were addressing by doing that?

[MicahChalmer]

I filed #40 to fix the typo and missing end of the comment.

As to @pnkfelix's question of why I made the mistake of setting syntax-begin-function in #30 in the first place...when I wrote #30 I was trying to translate from the previous approach of using syntax-propertize-function and syntax-propertize-extend-region. I was under the mistaken impression that syntax-begin-function was used in a similar way to the first element of the return value syntax-propertize-extend-region--that is, I thought it was used to start further back and thus avoid starting intepretation in the middle of a construct. But it's actually used to optimize--if syntax-begin-function is defined, it's only used to start further forward than the default, rather than further back. It's only there to make it go faster. (There's a check in syntax.el that ignores the return value if it's further back than the default.) Unless there are performance problems (and nobody has reported any so far) there is no reason to set it at all.

Of course, since I was also the one who put in the syntax-propertize-extend-region implementation as well, this naturally leads to follow-up questions of whether that was necessary (and why I thought it was.) I don't remember the answers to these.

I didn't put the above in the comments in #40--I figured it's a bit more than needed for ongoing maintenance of the code. Let me know if you think it should go in anyway, and I can add it as another commit to #40.

[vhbit]

@MicahChalmer I haven't chance to work a lot on Rust today, but on a quick look everything was fine, thanks!

[hugoduncan]

It seems to have fixed things for me. Many thanks for tracking that down.