BUG: port doesn't release (any?) memory

[flob@…]

After doing a selfupgrade from 1.320 to 1.442 and doing a upgrade installed after that in a new port-shell i saw port eating up my memory while doing noting than providing its shell after it finished its process of updating.

  UID   PID  PPID CPU PRI NI      VSZ    RSS WCHAN  STAT  TT       TIME COMMAND
    0 29148  1711   0  11 20  1331116 804640 -      SN+   p2   11:09.87 /usr/bin/tclsh /opt/local/bin//port

Thats about 800 MB of memory for a shell??????

[markd@…]

See related ticket #12022.

[kballard (Lily Ballard)]

markd, you do realize that you linked this very ticket, right?

[kballard (Lily Ballard)]

It's unfortunate that there aren't really any tcl debugging tools. There's no way (that I know of) to figure out what's going on here without random trial and error.

[jmpalacios (Juan Manuel Palacios)]

Milestone Needs developer review deleted

[tobypeterson]

Milestone MacPorts base bugs deleted

[jmroot (Joshua Root)]

I've narrowed down the memory usage problems to the filemap code. It adds about 160 MB to port's RPRVT each time it is used on my system. To demonstrate, open up port in interactive mode, run the activate, deactivate, provides or contents target, and watch your free memory disappear.

Note that it's far worse on trunk because it now closes the filemap when it's done with it (so more memory gets allocated next time it's opened). Upon closing, the variable is unset, FreeFilemapInternalRep() is called, and the ref count goes to zero, but no memory seems to be freed up.

[pguyot (Paul Guyot)]

Congrats on narrowing the bug to filemap code. Indeed, there was a memory leak there, which I fixed in r51419.