GitLab

The openFileDescriptor function is not user in the WIN32 codepaths,
so just #ifdef it out on WIN32.

First try to atomically open the file using O_CLOEXEC for open()
and the "e" mode for fopen(), and if that doesn't work or O_CLOEXEC
isn't defined, fall back to opening the file first and applying
the FD_CLOEXEC flag afterwards.

This is in preparation to making the wrapper enforce the O_CLOEXEC flag.

Use the openFile wrapper instead of calling fopen directly
in the libraries.

It was causing regressions for some half-broken files

Rationale:

Even if xref::constructXRef is returning false, that only means "i
din't find a new xref root", but constructXRef is also quite stubborn in
which it basically parses the whole file and stores the positions of all
the objects it finds, so for some half broken pdf files this is good
enough and we will be able to render those files just fine

At that point xref can have gone wrong since extractPDFSubtype() can
have caused a reconstruct that broke stuff so instead of unconditionally
returning true, return xref->isOk()

Fixes #706

SplashXPathScanner is only used inside the each fill function, but newer
passed to the outside. As it is small, there is no reason to not
allocate it on the stack.

Again no (C)

No (C) for this (a machine could do it)

Avoid global display profile state becoming an uncontrolled memory leak by enforcing single initialization. Closes #654

Do not try to parse into unallocated XRef entry and return pointer to dummy entry instead. Closes #692 and oss-fuzz/12330

Fixes #704

fixes oss-fuzz/12243

oss-fuzz/8548

Is a greeeeeeeeeeeeeat disk space saver

The CIDToGID map is an array of ints.  The code properly allocated
a number N of ints, but then used memcpy for N unsigned shorts.
That left the upper half of the array uninitialized.

This method was added in cc43c720
for the use in the ArturOutputDev::updateFont.
It is not use there anymore, and can therefore be removed again.

Previously, Splash code was used to get the codeToGID
mapping for font rendering (which you apparently cannot
get from Qt).  This patch reimplements the same functionality
using only FoFi and FreeType.  This makes the code easier
to understand, because it removes several layers of
redirection.

We provide both the build and runtime version, so if an app was built
against poppler 0.65 but running against 0.72, the app can say somewhere
in it's about/config dialogs

Running poppler 0.72 (built against 0.65)

This way we can explain to users while some features are not available
even if they should be, because the runtime and build time differences
sometimes matter