eric/Crosspoint
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Avoid OOM diring HTML parsing by reducing max block size and freeing unneeded buffers

Browse Source
This commit is contained in:
spfenwick
2026-05-02 16:07:27 +12:00
parent 913b7e5f8b
commit 5928f32364
2 changed files with 14 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files
lib/Epub/Epub/parsers/ChapterHtmlSlimParser.cpp
+6 -5
View File
@@ -1207,11 +1207,12 @@ void XMLCALL ChapterHtmlSlimParser::characterData(void* userData, const XML_Char
self->partWordBuffer[self->partWordBufferIndex++] = s[i];
}
// If we have > 750 words buffered up, perform the layout and consume out all but the last line
// There should be enough here to build out 1-2 full pages and doing this will free up a lot of
// memory.
// Spotted when reading Intermezzo, there are some really long text blocks in there.
if (self->currentTextBlock->size() > 750) {
// Flush when words approach the doubling boundary that would require a
// large contiguous realloc. 96 fires before capacity reaches 128
// (the next doubling after 64), keeping the realloc below 1.5KB and
// releasing excess capacity via shrink_to_fit in layoutAndExtractLines.
// The original 750-word threshold was too late for low-heap devices.
if (self->currentTextBlock->size() > 96) {
LOG_DBG("EHP", "Text block too long, splitting into multiple pages");
const int horizontalInset = self->currentTextBlock->getBlockStyle().totalHorizontalInset();
const uint16_t effectiveWidth = (horizontalInset < self->viewportWidth)