[Bug 1585] Allow an `Include' option which reads another config file in place and does not error out when `Include' file not readable

[bugzilla-daemon at bugzilla.mindrot.org]

https://bugzilla.mindrot.org/show_bug.cgi?id=1585

--- Comment #23 from Jakub Jelen <jjelen at redhat.com> ---
(In reply to Damien Miller from comment #22)
> There were a few problems with the previous patches, most due to
> host/match state persisting between files in non-intuitive ways.

This was partially intention. There is nothing worse than including
file into some unknown context and when the file returns, getting
different context. But in this seems to be solved in pretty clever way
and also tested in your current patch.

> This diff covers all the cases that I can think off and adds a
> regression test.

Thank you for having a look into that. I finally got to try that and it
seems to be working as expected (both regress tests and hand-checking).

Some thoughts/bugs:

 * if glob does not match any file, there should be at least
debug/verbose log about that. Certainly not fatal (as the bug title
proposes), but at least for debugging purposes. Something like this:

    verbose("%s: glob %s does not match any file", __func__, arg2);

 * in the last regress test, the include line is commented out, which
makes last tests pass regardless the included files:

@@ -332,7 +333,7 @@
-+#Include $OBJ/ssh_config.i.*
++Include $OBJ/ssh_config.i.*

-- 
You are receiving this mail because:
You are watching the assignee of the bug.
You are watching someone on the CC list of the bug.