Re: BLODA detection code in latest snapshot

[Ryan Johnson <ryan dot johnson at cs dot utoronto dot ca>]

On 29/02/2012 10:01 AM, Corinna Vinschen wrote:
> On Feb 29 09:51, Ryan Johnson wrote:
> > On 27/02/2012 7:26 AM, Corinna Vinschen wrote:
> > > Hi folks,
> > >
> > >
> > > I've just uploaded a new snapshot "2012-02-27 12:04:23 UTC".  It
> > > contains two code snippets which are supposed to help diagnosing BLODA
> > > problems.
> > >
> > > If you set the environment variable CYGWIN to "detect_bloda" and then
> > > start a Cygwin process (bash or so), then Cygwin will detect two types
> > > of anomalies:
> > > [...]
> > Would it be a good idea to update the FAQ's bloda entry with this
> > info? Sure, it's probably going to give occasional false positives
> > and/or negatives, but it would definitely catch the obvious cases
> > and give a quick test for claims of bloda-free systems. You'd almost
> > want a new cygcheck -b option that could fork off a process or two
> > with detect_bloda active and capture any output that results.
> Of course I will document this at one point.  So far I just didn't.
> I doubt that the cygcheck -b would be useful, though.  Just call
>
>    $ export CYGWIN=detect_bloda some_executable
>
> and you get what you want.
Sure. That's what I'd do also, but we're both familiar with the bloda. I 
was thinking more of users sending problem reports. Telling them to 
attach the output of `cygcheck -svrb' would give us useful information 
even if they don't (yet) know what the bloda is let alone whether 
they're affected by it.  Sort of like how we could ask users having 
strange problems to check for a second cygwin1.dll in their path... but 
it's easier to just ask for cygcheck output and check that. As a bonus, 
it would catch times where someone (*cough* me *cough*) thinks they're 
bloda free and so doesn't check for it before reporting a problem.

Heck, if we really wanted to go whole-hog, we could add an option to 
check for dlls in $PATH that have base collisions. Once cygcheck 
supported both those checks, the fork failure error message could even 
tell users to run cygcheck before reporting a problem.

Actually, now that I think about it, we could just make cygwin list any 
base collisions among dlls used by a failed forkee and point to the FAQ 
entry on rebaseall. The info is at our fingertips (dll::preferred_base) 
and in the absence of base collisions we could spawn a process to check 
for bloda, whose output (if non-empty) is highly likely to be relevant. 
The latency of a single spawn is nothing compared to ten failed fork 
attempts, so it wouldn't make cygwin any slower. Between those two 
checks, an intelligent user could deal with the vast majority of fork 
failures without ever invoking the mailing list. No change to cygcheck 
needed at that point.

Of course, SHTDI and I won't be able to until the semester ends, but it 
should be just a few hours' work.

Ryan

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple