GitLab

...

replace c with cc - glib

missing includes

oss-fuzz/27915

Fixes issue #997

Adobe Reader seems to be particularly strict about this

Issue #993

After playing hand editing files and opening them in Adobe Reader i
*think* the condition is "if the appearance stream has a ExtGState in
its Resources dict, then use the appearance stream, otherwise draw it
ourselves

StreamPredictor::getNextLine when predictori == 2 && nBits == 1 && nComps == 1

Issue #976
Issue #567

oss-fuzz/27619

When handling annotations without appearance stream, the behavior
of poppler deviated from what the pdf spec says:  Poppler would
only take the boundary width from a border style (BS) dictionary
if the 'style (S)' field was also present, even though the spec
clearly says that both are optional, and does not mention
one depending on the other.

This behavior was deliberate, because apparently Acroread 8
did it that way.  See the comment by Jeff Muizelaar in 28967940.
But it seems that Acroread behavior has changed, newer versions
do take the 'width' field into account even when there is no
'style (S)' field.  The Chromium pdf renderer does the same.

So let's change the code back to following the spec rather
than an old version of Acroread.

BUG: 988

If a line segment goes beyond the current clip region, Splash::strokeNarrow
tries to cut off the segment line to the intersection of the segment line
with the clip border.

Therefore it has to calculate a new endpoint (=intersection point) of the
segment. It does this by using the known segment slope dxdy.

Required information (segment start, segment slope, clip border) is known
in double precision. However the calculation used the integerized clip
border, which necessarily suffers from rounding errors. This error can
become very visible when we have a high dxdy (i.e, a flat-angle line).

Use SplashClip::getYMin and SplashClip::getYMax instead of getYMinI and getYMaxI,
and we get visibly correct results.

Fixes #990.

On some broken files m is 0 so every time we call
SampledFunction::transform we just return cachedOut, but it was never
initialized. We can solve it other ways but seems seems the least
expensive

Doesn't seem to regress any of the valid files i have and saves some
very small looping for hours

Simplifies having to remember to add the delete on all those error cases

Code looks better now

i.e. if the font name is DDPJAD+Times-Roman look for a replacement of Times-Roman

Fixes issue #972

Because PDF spec has no built in circle operator, we can only approximate
a circle by joining cubic bezier curve segments.

We already did good approximation for line endings and radio buttons by
setting distance of bezier control point to (4 * (sqrt(2) - 1) / 3) * r.
But the older AnnotGeometry::draw code for subtype circle still used
an equivalent of 0.5 * r, which yields a visibly non-circular circle.
Fix it by using the bezierCircle define for AnnotGeometry too.

Does minor refactoring by introducing AnnotAppearanceBuilder::drawEllipse,
and forwarding drawCircle to it. typeSquare code is slightly touched as
a consequence, but remains functionally unchanged.

Also changes 'S' (stroke path) to 's' (close and stroke path). To my
understanding 'close and stroke' is the correct thing to do here,
it's in line with drawCircle and the original commit 6c4fa513 didn't
explain why it left the path open by using 'S'.

Fixes #989.

poppler_add_unittest was only used 3 times & effectively differed from
poppler_add_test by just a CMake add_test call.

Only used once to check pkgconfig >= 0.18 (released in 2005 at latest).

Only used once & no longer needed; CMake configure_file converts
True/False to def/undef transparently now.

They are valid C++ but it's not what this project uses, it's a matter of
uniformity (same reason we have clang-format for example)

oss-fuzz/19177

Issue #979

It's were it makes sense, not as part of GlobalParams that we're
sloooooooooowly trying to get rid of