[RFC] Fix for GIO detection/VFS fallback (TCK7 failure)

[Andrew Hughes]

----- Original Message -----
> * Deepak Bhole <dbhole at redhat.com> [2012-05-02 12:12]:
> > * Andrew Hughes <ahughes at redhat.com> [2012-05-02 11:35]:
> > > ----- Original Message -----
> > > > ----- Original Message -----
> > > > > Hi,
> > > > > 
> > > > > Attached patch fixes a TCK failure seen on systems with older
> > > > > libgio.
> > > > > 
> > > > > The way the code works is that it first tries to load
> > > > > libgio-2.0.so
> > > > > and
> > > > > then verify that all necessary functions are available. If
> > > > > they are
> > > > > not,
> > > > > it dlcloses the library and falls back to vfs.
> > > > > 
> > > > > In the forest, we have modified this part to wire in system
> > > > > libs
> > > > > and
> > > > > there is a bug in how it is done. If functions are not found
> > > > > and
> > > > > the
> > > > > library handle is closed, the caller ignores it and tries to
> > > > > call a
> > > > > function (g_type_init) via the ptr anyway, causing a SIGSEGV.
> > > > > 
> > > > > OK to apply this to HEAD, 2.0 and 2.1?
> > > > > 
> > > > > Thanks,
> > > > > Deepak
> > > > > 
> > > > 
> > > > I saw this earlier and was going to commit it to HEAD.
> > > > 
> > > > Fine by me.  I tested with and without system GIO enabled, but
> > > > I
> > > > can't
> > > > really emulate the library being missing without deleting
> > > > chunks of
> > > > my
> > > > system... ;-)
> > > > 
> > > > We need to find time to get these patches upstreamed.
> > > > --
> > > > Andrew :)
> > > > 
> > > > Free Java Software Engineer
> > > > Red Hat, Inc. (http://www.redhat.com)
> > > > 
> > > > PGP Key: 248BDC07 (https://keys.indymedia.org/)
> > > > Fingerprint = EC5A 1F5E C0AD 1D15 8F1F  8F91 3B96 A578 248B
> > > > DC07
> > > > 
> > > > 
> > > 
> > > FYI, I put this into HEAD prior to the 2.2 branch.  I'll leave
> > > you
> > > to handle 2.0 and 2.1.
> > 
> > Sure. Also, this does not affect upstream btw. The upstream code
> > correctly detects when libgio is inadequate and falls back. The bug
> > was
> > introduced by this changeset:
> > http://icedtea.classpath.org/hg/icedtea7-forest/jdk/rev/8317cb1976d6
> > 
> 
> Unless you were referring to getting the SYSTEM_GIO patch itself
> upstream, in which case, nvm :)

I was :-)

I know where the bug comes from; I wrote the system GIO usage and thus
the bug... it's unfortunately not something you can reproduce if you do
have an up-to-date GIO installed.

I am surprised jtreg tests haven't hit this earlier though.  Do none test this code?

> 
> Deepak
> 
> > Cheers,
> > Deepak
> > 
> > > Cheers,
> > > Roman
> > > 
> > 
> > > --
> > > Andrew :)
> > > 
> > > Free Java Software Engineer
> > > Red Hat, Inc. (http://www.redhat.com)
> > > 
> > > PGP Key: 248BDC07 (https://keys.indymedia.org/)
> > > Fingerprint = EC5A 1F5E C0AD 1D15 8F1F  8F91 3B96 A578 248B DC07
> > > 
> 

-- 
Andrew :)

Free Java Software Engineer
Red Hat, Inc. (http://www.redhat.com)

PGP Key: 248BDC07 (https://keys.indymedia.org/)
Fingerprint = EC5A 1F5E C0AD 1D15 8F1F  8F91 3B96 A578 248B DC07