Re: bug: Useless error message when SVN_SSH points to non-existent file

[Greg Hudson <gh...@MIT.EDU>]

On Fri, 2003-06-06 at 20:49, Garrett Rooney wrote:
> Unfortunately, it seems that fixing this will require changes to APR.

I don't like the idea of trying to find out in advance whether the
tunnel agent is executable.  There's no completely robust way of doing
so under Unix (access() runs afoul of setuid, eaccess() isn't portable,
and checking permission bits by hand doesn't work right with AFS), and
it's complicated at any rate because of the path search.

One option is to change the protocol so that the "greeting" can be an
error message.  That change could have other applications down the road
("this repository has moved").  We can either do this Very Soon (before
0.24) in a backward-incompatible manner, by changing the format of the
greeting, or we can do it later in a backward-compatible but less
elegant manner, by compatibly extending the greeting syntax.  I will
look into doing it the first way now, I think; if I run into problems,
we can always do it the other way later.

(From an APR perspective, we can use apr_procattr_child_errfn_set to
write the error message when the exec doesn't succeed.)


---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: bug: Useless error message when SVN_SSH points to non-existent file

[Garrett Rooney <ro...@electricjellyfish.net>]

On Friday, June 6, 2003, at 10:48 PM, Greg Hudson wrote:

> On Fri, 2003-06-06 at 20:49, Garrett Rooney wrote:
>> Unfortunately, it seems that fixing this will require changes to APR.
>
> I don't like the idea of trying to find out in advance whether the
> tunnel agent is executable.  There's no completely robust way of doing
> so under Unix (access() runs afoul of setuid, eaccess() isn't portable,
> and checking permission bits by hand doesn't work right with AFS), and
> it's complicated at any rate because of the path search.
>
> One option is to change the protocol so that the "greeting" can be an
> error message.  That change could have other applications down the road
> ("this repository has moved").  We can either do this Very Soon (before
> 0.24) in a backward-incompatible manner, by changing the format of the
> greeting, or we can do it later in a backward-compatible but less
> elegant manner, by compatibly extending the greeting syntax.  I will
> look into doing it the first way now, I think; if I run into problems,
> we can always do it the other way later.
>
> (From an APR perspective, we can use apr_procattr_child_errfn_set to
> write the error message when the exec doesn't succeed.)

Wow, that's quite a bit cooler than the way I was going, thanks Greg ;-)

-garrett


---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org