lvfntman: fix harfbuzz related memory leak on quit

[benoit-pierre]

This change is 

[poire-z]

What's the reason for moving the check ==NULL inside the function (so, always calling this function that does its work only once) - rather that doing the check here and really calling this function once ?
(This is called for each text segment in SVGs.)

[benoit-pierre]

It's cleaner that way IMHO. I'm not concerned about performance (but I've made setup_SVGGlyphsCollector_svg_funcs static so the compiler can do its thing). (Note that I do benchmark regularly to avoid regressions).

[poire-z]

It's cleaner that way IMHO

I don't feel the same: one has to go jump at the function to see it's a no-op on next calls.
And I don't trust compilers to intuite the logic I thought about but was lazy to write :)

[poire-z]

"memory leak on quit" sounds like the best memory leaks :)
Why chase them ? Just to get them out of the way to catch the other ones? Or some other reason?

[NiLuJe]

It makes Valgrind reports tedious to parse, making finding meaningful leaks that much more annoying ;).

[benoit-pierre]

"memory leak on quit" sounds like the best memory leaks :) Why chase them ? Just to get them out of the way to catch the other ones? Or some other reason?

Well, technically, leaks on dlclose.

It makes Valgrind reports tedious to parse, making finding meaningful leaks that much more annoying ;).

This.

I'm down to the expected 32 bytes (freetype memory object) in a lot of cases.