Open Radar

 
Area:
Core Graphics

Summary:

iOS 11 introduced a cache used by CGPDFDocumentGetPage, but it has unwanted performance implications.

When a new page is requested, it's added to the cache (CGPDFDocumentAppendMRUPage). The least recently used page is then probably removed and old images are purged (CGPDFPagePurgeImages)

To know which images can be purged, the old page is probably parsed again (CGPDFScannerScan), but for heavy PDF documents, this can be very CPU- and memory intensive.

The worst thing is, this even happens when only metadata is requested, e.g. looping over all CGPDFPageRefs in a document and requesting their document dictionary (CGPDFPageGetDictionary). In this case, CGPDFPagePurgeImages won't even need to do anything. 


Steps to Reproduce:

Open a complex multi-page PDF document
Loop over all pages, and get their page dictionaries


Expected Results:

Like on iOS 10, this should be a linear task, not relative to the complexity of the pages.


Actual Results:

A lot of memory and CPU is consumed for no obvious reason.


Version/Build:

iOS 11.0.1


Configuration:

Device and simulator