---- Reported by grompf@sublimeintervention.com 2005-09-07 21:33:57 MST ----

This might be a possible file handle leak in gmcs?  io-layer?

I'm leaning towards gmcs as mcs can compile the 1.1 corlib fine (and older gmcs's can compile SVN 
fine).

To reproduce: ulimit -n 256 (default on OSX)
Rebuild mcs from a clean state.

gmcs will report that it cannot find source files (FileNotFoundEx) when they do in fact exist.

ulimit -n 8192 and gmcs compiles corlib fine.



---- Additional Comments From miguel@ximian.com 2005-09-09 01:13:32 MST ----

Dick, can you look at this?  It seems like a potential leak



---- Additional Comments From grompf@sublimeintervention.com 2005-09-09 01:24:51 MST ----

miguel,

  As I said above, I'm fairly certain this isn't a leak in io-layer as mcs can compile the 
net_1_1_bootstrap corlib fine (with the bootstrapped mcs); but gmcs cannot compile the 
net_2_0_bootstrap corlib with the bootstrapped gmcs.  They use the same corlib.sources.dll 
meaning they will open the same number of files (unless I'm misunderstanding something 
drastically?)

-kangaroo




---- Additional Comments From grompf@sublimeintervention.com 2005-09-09 01:38:41 MST ----

Created an attachment (id=168476)
patch




---- Additional Comments From grompf@sublimeintervention.com 2005-09-09 01:39:52 MST ----

This isn't a bug for Dick; there is a FIXME commenting out a block in driver.cs and the 
input.Close () was missing leaving file handles lingering for the GC to pickup.  The attached 
patch fixes the problem.

Reassigning to mono-bugs for commit approval

-kangaroo




---- Additional Comments From grompf@sublimeintervention.com 2005-09-09 10:44:58 MST ----

fixed in SVN r49804

-kangaroo


Imported an attachment (id=168476)

Unknown operating system unknown. Setting to default OS "Other".