bug#67142: 29.1; with-sqlite-transaction commits on exception rather than rolling back

[Vasilij Schneidermann <mail@vasilij.de>]

[-- Attachment #1: Type: text/plain, Size: 981 bytes --]

> > I don't understand this part, sorry.  Why do we need a symbol to
> > return the result of the body?
> 
> db-var and func-var are uninterned symbols in the macro expansion but
> the variables 'result' and 'commit' marked below aren't.  If the intent
> is not to expose these variables to the macro's BODY, the inner let
> should also uninterned symbols right?

Correct, this is what I was hinting at. As annoying as it is, when
writing unhygienic macros one should use uninterned symbols to avoid
exposing additional variables to the BODY argument.

Alternatively, the pattern of `(let ((return (...))) ... return)` can be
replaced with `(prog1 (...) ...)`.

@Eli: The new patch looks better and closer to how the issue is solved
in the Ruby sqlite3 gem. I'm still not sure about this use of
`unwind-protect` being correct, but it does preserve the backtrace in
case of an error better than when I used `(condition-case e (...) (error
(apply #'signal e)))`.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]