| Age | Commit message (Collapse) | Author |
|---|
|
Some macOS users were experiencing build failures where pkg-config
could not find poppler, even with Homebrew's poppler package installed.
This occurs when Homebrew doesn't properly link the .pc files to the
standard pkg-config search path.
Fixes #299
|
|
Alpine Linux deprecated sudo in favor of doas. Add doas support to
exec_privileged() so autobuild works on Alpine without requiring users
to install and configure sudo.
Fixes #323
|
|
|
|
Notes from the original PR:
The dependencies for android in autobuild can not be installed in
pdf-tools-install, they have to be installed manually in Termux. But
the dependencies are checked when running autobuild.
pdf-tools-install successfully installs epdfinfo in Android Emacs. One
can then view pdf files.
The shebang in autobuild does not work in Termux, I therefore had to
create a separate script "autobuild.android" that in turn executes
autobuild. The problem with the shebang is that "env" in Termux is not
located at "/usr/bin/env".
pdf-tools/Makefile does currently not work in Android Emacs, but it is
not needed for running pdf-tools-install. Fixing this will have to be
done in a separate pull request, if deemed necessary. Reason: There is
yet no consensus for how to implement a command line emacs, to be run
in batch mode, in Android Emacs.
I have successfully run "make test" in Ubuntu, in the repo feature
branch. Running "make server-test" in the master branch fails, so I
have not tested this in the feature branch.
Closes: #230
Author: @johanwiden
|
|
Add gamma correction options for better contrast control in midnight mode:
- pdf-view-midnight-gamma: nonlinearly scale lightness (default 1.0)
- pdf-view-midnight-gamma-before-invert: control gamma application order
Bug fixes:
- Fix RGB normalization (255 not 256) - was causing incorrect background color
- Add missing black→foreground fast path (comment existed but code was missing)
Closes: #329
Author: @vizziv
|
|
Add build support for Nobara Linux, a Fedora derivative.
- Add os_nobara() detection function (checks /etc/nobara-release)
- Uses same packages as Fedora (dnf package manager)
- Add to --os argument handling and auto-detection chain
Closes: #301
Author: @gs-101
|
|
Use platform-appropriate temp file creation:
- _tempnam() on Windows (Microsoft-recommended)
- mkstemp() on Unix-like systems
Fixes #110, addresses #316
Supersedes #235
|
|
Support the newer UCRT64 MSYS2 environment (recommended default per msys2.org).
- Add UCRT64 detection to server/autobuild
- Fix vasprintf implicit declaration error on Windows (synctex_parser.c)
- Update configure.ac for Windows compatibility
Fixes #282, #286
Supersedes #295
Tested by multiple users on Windows 10/11 with MSYS2.
Co-authored-by: legends2k
|
|
Copying my comment from the Github issue:
> The problem is that something seems to have changed in the way brew
> installs glib, due to which pkg-config does not find the glib headers.
>
> To fix this, I have updated the autobuild script to manually modify
> the PKG_CONFIG_PATH. This should fix the problem.
>
> I will test the change and close this issue via the commit once I push
> it. You will need to update your version of pdf-tools and then restart
> Emacs to trigger the re-build again. (It might take a few hours for
> Melpa to pick up the latest version of the pdf-tools)
>
> I will create a linked issue to investigate a better long-term
> solution to this problem.
Closes: #269
|
|
|
|
This allows calling `pdf-info-addannot` from Lisp with a nil
`markup-edges` argument, which was previously disallowed.
In this case, the annotation covers exactly the selected region of the
page, without attempting to match it to a certain text region.
This means that you can now highlight an exact region! Just select the
region by holding down the Meta key and dragging the mouse. Then
invoke `pdf-annot-add-highlight-markup-annotation` (with `C-c C-a h`)
and voila!
Closes: #191
|
|
From the comments in the PR, @smithzvk explains the changes:
> I was bugged that it didn't pick up the hue from the background,
> however, so I made the all color a/b parameters start from the a/b
> parameters of the BG color. This makes the PDF background (white)
> map to your BG color and PDF foreground (black) map to your FG
> color, which can be set to your theme background making it seem
> pretty seamless. Note that if you set your BG and FG to a color
> theme without zero saturation (a gray tone of some sort), then you
> recover the original behavior.
>
> I also reinstated the cached transforms for background and the last
> color computed.
>
> I also refactored the loop to move the switch statement outside of
> the loop. It is more repetitive, but it was needed to get rid of
> some warnings about potentially uninitialized variables. It should
> also be faster as the branch was removed from the loop.
>
> I also ditched the gamma correction as it seemed to be doing the
> wrong thing and I never had a theoretical reasoning for it. I can
> get similar results if I just set my BG color to something like
> gray5 instead of black. Barely a difference in background but the
> hinting on the fonts is showing again. The root of my issue may
> have always been some kind of display calibration issue.
Closes: #197
|
|
Uses the Lightness values from the `pdf-view-midnight-colors` `fg` and
`bg` to scale the inverted color's lightness. Band-aid till better
method is found.
Relates to: #69
|
|
In Void Linux most heading/development files are separated from the "main" packages and are stored on `${package-name}-devel`. The autobuild for Void fails because it tries to find `cairo-devel`, which contains `cairo`, respectives headers and other dependencies.
This Pull Request adds `cairo-devel` to the `PACKAGES` variable
|
|
This inversion method attempts to maintain the color hue and saturation but
inverts the lightness using the OKLab color space[^1].
[^1]: https://bottosson.github.io/posts/oklab/
* server/epdfinfo.c (image-recolor): Add feature to support the OKLab
inversion method functionality
* lisp/pdf-view.el (pdf-view-midnight-invert): Add new variable to
invert the image color lightness while maintaining hue.
(pdf-view-midnight-minor-mode): Account for above.
* lisp/pdf-info.el (pdf-info-query--parse-response): Handle changes to
:render/usecolors command
Closes: #69
Closes: #169
Closes: politza/pdf-tools#698
Closes: politza/pdf-tools#608
|
|
Make selection-style configurable via `pdf-view-selection-style`. A
default value of `word` might be desirable, as selection is usually
intended to select entire words is perceptually faster than
glyph-based selection.
This affects selections and highlights, but not search.
Limitation:
The selection type is fixed for all currently selected regions.
Selecting multiple regions with different selection styles at the same
time would require storing the selection style for every region and
replacing the protocol for `cmd_addannot` with a more flexible one,
e.g. the keyword-based one used by `cmd_renderpage`.
Closes: #73
|
|
|
|
This commit is a squashed commit of PR #164 by @skyler544
Author: Skyler Mayfield <skyler544@gmail.com>
Date: Fri Oct 21 12:00:46 2022 +0200
Closes: #164
|
|
Closes: #170
|
|
Until this commit, docker-based local testing only checked whether the
epdfinfo server compiled correctly. With this commit, local testing
now also runs elisp tests against the different versions of Emacs we
currently support.
All versions of Emacs are run on Ubuntu at the moment,
but going forward I expect that we will add more operating system
variants to this matrix as well.
As test coverage increases, this will help automate away a lot of the
tests!
Relates to: #130
|
|
Also, simplify the config script for finding gnu-sed.
Relates to: #160
|
|
This ensures that all operating systems currently mentioned in the
README are covered in the autobuild script
Relates to: #160
|
|
This is pretty cool, I hope it works! 🤞 It uses the `macos` orb
provided by CircleCI, as documented here:
https://circleci.com/docs/using-macos
This commit also rewrites the older anchor style common steps into the
newer command syntax, which is more readable. The command syntax is
documented here:
https://circleci.com/docs/configuration-reference#commands
Installing pre-reqs on MacOS via `brew` was non-trivial. This is
mainly because of the `sed` command used in the `Makefile`. It isn't a
problem normally because it is only used for packaging, which is done
on Melpa
Relates to: #98
|
|
This is just a convenience wrapper for me
|
|
This allows re-using containers, saving a tremendous amount of time.
The drawback here is that any environment change (on the docker /
podman containers side) is not detected. The way to deal with this
will be to create a process around major releases where I delete all
the containers and re-test from scratch.
Relates to: #130
|
|
This reverts commit 0107f80dcf268d08ac075b01729820062c0fbd67
The reason for doing this is explained by @vizs in #143, copying the
message here:
Andreas wrote the function
`annotation_translate_quadrilateral`[^1][^2] to fix misplaced
annotations due to the then bug in `glib`. This misplacement bug can
be seen in the images posted in issue politza/pdf-tools#577, along
with PDFs in which the bug can be reproduced.
This lack of translation in the `glib` backend was fixed in `poppler
0.86.0` but the workaround was still left hanging in `epdfinfo.c`
which brings us to today where this unneeded translation bites our
back. I commented out the calls to
`annotation_translate_quadrilateral` in `epdinfo.c` and to my
surprise, there were no misplaced translations in files which
displayed the misplacing behaviour and in files which did not have
this behaviour.
[1]:
https://github.com/politza/pdf-tools/commit/0107f80dcf268d08ac075b01729820062c0fbd67
[2]: politza/pdf-tools#122
Closes: #143
Closes: #146
|
|
Recently nixpkgs removed the `pkgconfig` alias[^1], which broke the
build with `nix-shell`.
[^1]: https://github.com/NixOS/nixpkgs/pull/192681
Closes: #153
Closes: #158
|
|
(#134)"
This reverts commit bea5ddb9fc234b48db6df3dcb66d75e76bec00c8 because
`pdf-tools` compilation is broken.
Closes: #141
Reopens: #134
|
|
The vasprintf() is a GNU extension function, missed in non-GNU build environment, so we define it.
Without this change, there will have error message when we build with clang:
```
./autogen.sh
./configure CC=clang
make
```
We will get errors:
```
synctex_parser.c:8448:13: error: call to undeclared function 'vasprintf'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]
if (vasprintf(&buffer, format, va) < 0) {
```
This change fixes the problem.
|
|
* Use `--run` flag to execute nix-shell
The `--command` flag is only required to spawn a interactive shell.
* Do not require gcc to build with nix-shell
`nix-shell` already sets CC with a C compiler, for instance Clang on darwin.
* Remove outdated require in test-helper
`f.el` is not needed anywhere.
|
|
They aren't used.
Fixes #138.
|
|
The output on the terminal suggested using fix-missing, which seems to
fix the sources and find the packages :)
|
|
README changes:
- Add documentation on how to test / add support for your favorite
operating system.
Dockerfile changes:
- Rename Dockerfile.in to Dockerfile.common.in, to make the name
clearer.
- Add make as a prerequisite in the Dockerfile.
+ It is needed to run make commands and is not available in some of
these images.
- Add make targets for printing operating systems that are not yet
supported in Docker testing.
+ This should help contributors with adding support for testing on
these systems.
+ New targets are `server-test-supported` and
`server-test-unsupported` at the top level.
Related to: #130
|
|
|
|
No point leaving dead containers lying around.
Relates to: #98
|
|
This simplifies container operations and ensures that we don't have
dead containers hanging around.
Relates to: #98
|
|
- Move all package installation to the autobuild file.
The Dockerfile should simply start a container and make sure that
all the package repositories are updated.
- Dockerfiles should spin up containers for the latest (ish) releases
of the major Linux operating systems.
- Disable Dockerfiles for environments which are not working
correctly.
List of operating systems where `epdfinfo` server compilation is
tested and working correctly:
- Debian (9, 10, 11)
- Ubuntu (18, 20, 22)
- Fedora (34, 35, 36)
List of operating systems where `epdfinfo` server compilation is
failing inside the containers, for unknown reasons:
- Gentoo
- Centos
- Archlinux
List of operating systems that need to be added to containerized
testing:
- MacOSX
- Windows via Msys2 (Mingw32, Mingw64)
- OpenBSD
- FreeBSD
- NixOS
- Void
- OpenSuse
Relates to: #98
|
|
This commit also removes C++ entirely, as it is not needed anymore for
correct compilation of `epdfinfo`. The following changes are made:
- Remove private headers from all the server code. This involves:
- Replace private header functionality with the correct public
functions provided by poppler.
- Remove unneeded headers around date checks
- Delete dead-code (headers, m4 macros)
- Remove C++, g++ and similar aliases from all the code. Also:
- Remove related information from the README
- Remove related information from CI and Dockerfiles
Work done by: @Pi-Cla
Closes: #105, #103, #109, #114
|
|
This reverts commit 37bbe861755bc60c7cc333359fee3e2a5d919c77.
As explained by @ShuguangSun (re: testing the change on Windows) on
> Sorry. It is my fault. The issue is still there. The different is
> that the temp dir is created now. However, it reports "Unable to
> create temporary file".
Reopens: #110
|
|
This reverts commit 8ee31220a6ae3e41549bfffca7a89c481d270004 and
brings in @JunyuanChen's change once again.
We modify the commit to fix the problem on Windows by changing the
code as follows:
--- a/server/epdfinfo.c
+++ b/server/epdfinfo.c
@@ -347,6 +347,6 @@
static char*
mktempfile()
{
- char template[] = "/tmp/epdfinfoXXXXXX";
+ char template[] = P_tmpdir "/epdfinfoXXXXXX";
char *filename = malloc(sizeof(template));
memcpy(filename, template, sizeof(template));
This works correctly for Windows (as confirmed by @ShuguangSun),
Mac (tested by @vedang) and Linux (tested by @Junyuanchen)
Relates to: #101, #94
Closes: #110
|
|
This reverts commit d63a1e7d87f9b0a19209f2eeb170bcf64612aa2f.
In d63a1e7, @JunyuanChen fixed a long standing compilation warning by
replacing `tempnam` with `mkstemp` in `epdfinfo`. However, the
`mkstemp` implementation does not work correctly on MS
Windows (probably because the path template is "wrong" for Windows).
I am reverting the commit and opening a new issue #110 to track the
correct implementation of `mkstemp` (or equivalent) on MS Windows.
It would also be great to add a test against Windows CI to open a PDF
and check that the operation completes successfully.
Closes: #101
|
|
As explained in #102 by @Pi-Cla and @uliw, `pdf-tools` relies on
private headers provided by `poppler`. In order to compile these
correctly, we need the standard library provided by C++17.
`pdf-tools` is built and tested against C++11. This workaround
provides the correct CXXFLAGS to compile the library and fixes the
immediate broken compilation problem.
There should be two "correct" fixes to these:
1. Remove dependence on private `poppler` headers. This is being
tracked in #103 and #105
2. Upgrade `pdf-tools` to build and run against new C++ versions. This
is being tracked in #109.
Closes: #70, #76, #102, #108, politza/pdf-tools#706.
|
|
Nearly forgot to add pkgconf.
|
|
I am not qualified to add support for the Micros versions of openSUSE, but this supports the most commonly used versions.
|
|
This is a better solution as it takes into account paths that have
already been set up by users.
Potentially fixes: #70
|
|
This fixes the `server/autobuild` installation on Windows, as
mentioned in #86.
Fixes: #86
|
|
Removes the following warning when compiling `epdfinfo`:
```
build/server/epdfinfo.c:354: warning: the use of `tempnam' is
dangerous, better use `mkstemp'
```
|
|
People who use [Nix Flakes](https://nixos.wiki/wiki/Flakes) do not
usually have `nixpkgs` in their `NIX_PATH`, which leads to failed
builds, because `nix-shell` relies on it. I suppose some users might
not have `nixpkgs` in `NIX_PATH` for other reasons also. I propose to
use [NixOS/nixpkgs](https://github.com/NixOS/nixpkgs) in such cases
Merges, Closes: #72
|
|
Since 6th Jan, 2022, builds on Appveyor are failing because they
cannot find `autoreconf`[1][2].
This should be implemented from `base-devel`, but even with
`base-devel` available the builds are failing. Adding an explicit
require on `autoconf` in the hope that this is fixed.
1: https://ci.appveyor.com/project/vedang/pdf-tools/builds/42114287
2: https://ci.appveyor.com/project/vedang/pdf-tools/builds/42101965
Relates to: #68
|
|
Very simple addition to the autobuild script to include the xbps
package manager and Void Linux.
Closes: politza#685
|
