Re: gitk: avoid obscene memory consumption

Am 07.11.2016 um 05:11 schrieb Paul Mackerras:

So far I've found three arrays being populated lazily (which is good) but never being released (which ignores changes to the underlying repo):

$commitinfo: one entry of about 500 bytes per line viewed in the list of commits. Maximum size of the array is the number of commits. As far as I can see, this array should be removed on a reload (Shift-F5).

$blobdifffd: one entry of about 45 bytes for every commit ever read. The underlying file descriptor gets closed, but the entry in this array remains. So far I didn't find the reason why this array exists at all. It's also not removed on a reload.

$treediffs: always the same number of entries as $blobdiffd, but > 1000 bytes/entry. Removed/refreshed on a reload (good!), different number of entries from that point on.

In case you want to play as well, here's the code I wrote for the investigation, it can be appended right at the bottom of the gitk script:

--------------8<---------------
proc variableSizes {} {
    # Add variable here to get them shown.
    global diffcontext diffids blobdifffd currdiffsubmod commitinfo
    global diffnexthead diffnextnote difffilestart
    global diffinhdr treediffs

    puts "---------------------------------------------------"
    foreach V [info vars] {
	if { ! [info exists $V] } {
	    continue
	}

	set count 0
	set bytes 0
	if [array exists $V] {
	    set count [array size $V]
	    foreach I [array get $V] {
		set bytes [expr $bytes + [string bytelength $I]]
	    }
	} elseif [catch {llength [set $V]}] {
	    set count [llength [set $V]]
#	    set bytes [string bytelength [list {*}[set $V]]]
	} else {
	    set bytes [string bytelength [set $V]]
	}
	puts [format "%20s: %5d items, %10d bytes" $V $count $bytes]
    }

# catch { # set output [memory info] # puts $output # }

    after 3000 variableSizes
}

variableSizes -------------->8---------------

[memory info] requires a Tcl with memory debug enabled.

Markus