On Mar 5, 2007, at 09:51, Abel Braaksma wrote:
Andreas L Delmelle wrote:
AFAICT, it is already parsed with SAX (see
org.apache.fop.fonts.FontReader).
Perhaps it is something else then? :S
It doesn't seem like it's those widths by itself that are causing
the problem. An array of 50K int-widths should correspond to min.
200K max. 400K of heap. :/
That sounds logical.
You mentioned that the OOMError in Tomcat occurred right after
loading the metrics, which would indicate that it's actually what
follows that's taking up the memory.
I'm no expert on the font-handling, but browsed around through the
related code, and found that at some point the entire TTF file gets
read into a byte array
(see org.apache.fop.fonts.truetype.FontFileReader()).
[Note:
If I judge correctly, there is even the possibility that this happens
twice? If there is no metrics file, then LazyFont.load() does this
first, and later on, an independent FontFileReader() is used by the
renderer to embed the font... I'll leave this for others to comment in.
Also, FontFileReader.init() transfers the file contents to a byte
array but the constructor does not close the InputStream. I wonder if
that's intended or an oversight, especially since the other
constructor does close the stream?]
Anyway, while this decreases time spent on I/O later on, for such
large fonts this does indeed look like a suboptimal approach...
especially since ultimately only the used glyphs get embedded in the
result. :/
Cheers,
Andreas
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
