qpdf/README-maintainer

ROUTINE DEVELOPMENT

**Remember to check pull requests as well as issues in github.**

Default:

./configure CXX="g++ --std=c++14" --enable-werror --enable-maintainer-mode

Debugging:

./configure CXX="g++ --std=c++14" CFLAGS="-g" CXXFLAGS="-g" \
   --enable-werror --disable-shared --enable-maintainer-mode

Profiling:

./configure CXX="g++ --std=c++14" CFLAGS="-g -pg" CXXFLAGS="-g -pg" \
   LDFLAGS="-pg" --enable-werror --disable-shared --enable-maintainer-mode

Then run `gprof gmon.out`. Note that gmon.out is not cumulative.

Memory checks:

./configure CFLAGS="-fsanitize=address -fsanitize=undefined -g" \
   CXXFLAGS="-fsanitize=address -fsanitize=undefined -g" \
   LDFLAGS="-fsanitize=address -fsanitize=undefined" \
   CC=clang CXX="clang++ --std=c++14" \
   --enable-werror --disable-shared --enable-maintainer-mode


VERSIONS

* The version number on the main branch is whatever the version would
  be if the top of the branch were released. If the most recent
  release is version a.b.c, then

  * If there are any ABI-breaking changes since the last release,
    main's version is a+1.0.0
  * Else if there is any new public API, main's version is a.b+1.0
  * Else if there are any code changes, main's version is a.b.c+1

* Whenever we bump the version number, bump shared library versions as
  well.

* Every released major/minor version has an a.b branch which is used
  primarily for documentation but could potentially be used to create
  a new patch release after main has moved on. We don't do that as a
  rule, but there's no reason we couldn't do it if main had unreleased
  ABI/API changes that were still in flux and an important bug fix was
  needed on the most recent release. In that case, a release can be
  cut from a release branch and then either main can be rebased from
  there or the changes can be merged back, depending on the amount of
  drift.


CHECKING DOCS ON readthedocs

To check docs on readthedocs.io without running all of CI, push to the
doc-check branch. Then visit https://qpdf.readthedocs.io/en/doc-check/
Building docs from pull requests is also enabled.


GOOGLE OSS-FUZZ

* See ../misc/fuzz (not in repo) for unfixed, downloaded fuzz test cases

* qpdf project: https://github.com/google/oss-fuzz/tree/master/projects/qpdf

* Adding new test cases: download the file from oss-fuzz and drop it
  in fuzz/qpdf_extra/issue-number.fuzz. If not ready to include, it
  can be stored anywhere, and the absolute path can be passed to the
  reproduction code as described below.

* To test locally, see https://github.com/google/oss-fuzz/tree/master/docs/,
  especially new_project_guide.md. Summary:

  Clone the oss-fuzz project. From the root directory of the repository:

  Add `-e GITHUB_FORK=fork -e GITHUB_BRANCH=branch` to build_fuzzers
  from a qpdf fork/branch rather than qpdf/main.

  python3 infra/helper.py build_image --pull qpdf
  python3 infra/helper.py build_fuzzers [ --sanitizer memory|undefined|address ] qpdf
  python3 infra/helper.py check_build qpdf
  python3 infra/helper.py build_fuzzers --sanitizer coverage qpdf
  python3 infra/helper.py coverage qpdf

  To reproduce a test case, build with the correct sanitizer, then run

  python3 infra/helper.py reproduce qpdf <specific-fuzzer> testcase

  where fuzzer is the fuzzer used in the crash.

  The fuzzer is in build/out/qpdf. It can be run with a directory as
  an argument to run against files in a directory. You can use

  qpdf_fuzzer -merge=1 cur new >& /dev/null&

  to add any files from new into cur if they increase coverage. You
  need to do this with the coverage build (the one with
  --sanitizer coverage)

* General documentation: http://libfuzzer.info

* Build status: https://oss-fuzz-build-logs.storage.googleapis.com/index.html

* Project status: https://oss-fuzz.com/ (private -- log in with Google account)

* Latest corpus:
  gs://qpdf-backup.clusterfuzz-external.appspot.com/corpus/libFuzzer/qpdf_fuzzer/latest.zip


CODING RULES

* In a source file, include the header file that declares the source
  class first followed by a blank line. If a config file is needed
  first, put a blank line between that and the header followed by
  another blank line. This assures that each header file is included
  first at least once, thereby ensuring that it explicitly includes
  all the headers it needs, which in turn alleviates lots of header
  ordering problems. The blank line ensures that formatters don't
  mess this up by resorting the headers.

* Avoid atoi. Use QUtil::string_to_int instead. It does
  overflow/underflow checking.

* Avoid certain functions that tend to be macros or create compilation
  errors on some platforms. Known cases: strcasecmp, abs. Avoid min
  and max. If needed, std::min and std::max are okay to use in C++
  code with <algorithm> included.

* Remember to avoid using `operator[]` with `std::string` or
  `std::vector`. Instead, use `at()`. See README-hardening.md for
  details.

* Use QIntC for type conversions -- see casting policy in docs.

* Remember to imbue ostringstreams with std::locale::classic() before
  outputting numbers. This protects against the user's global locale
  altering otherwise deterministic values. (See github issue #459.)
  One could argue that error messages containing numbers should
  respect the user's locale, but I think it's more important for
  output to be consistent, since the messages in question are not
  really targeted at the end user.

* Use QPDF_DLL on all methods that are to be exported in the shared
  library/DLL. Use QPDF_DLL_CLASS for all classes whose type
  information is needed. This is important for exception classes and
  it seems also for classes that are intended to be subclassed across
  the shared library boundary.

* Put private member variables in PointerHolder<Members> for all
  public classes. Remember to use QPDF_DLL on ~Members(). Exception:
  indirection through PointerHolder<Members> is expensive, so don't do
  it for classes that are copied a lot, like QPDFObjectHandle and
  QPDFObject.

* Traversal of objects is expensive. It's worth adding some complexity
  to avoid needless traversals of objects.

* Avoid attaching too much metadata to objects and object handles
  since those have to get copied around a lot.


HOW TO ADD A COMMAND-LINE ARGUMENT

QPDFJob is documented in three places:

* This section provides a quick reminder for how to add a command-line
  argument

* generate_auto_job has a detailed explanation about how QPDFJob and
  generate_auto_job work together

* The manual ("QPDFJob Design" in qpdf-job.rst) discusses the design
  approach, rationale, and evolution of QPDFJob.

Command-line arguments are closely coupled with QPDFJob. To add a new
command-line argument, add the option to the appropriate table in
job.yml. This will automatically declare a method in the private
ArgParser class in QPDFJob_argv.cc which you have to implement. The
implementation should make calls to methods in QPDFJob via its Config
classes. Then, add the same option to either the no-json section of
job.yml if it is to be excluded from the job json structure, or add it
under the json structure to the place where it should appear in the
json structure.

In most cases, adding a new option will automatically declare and call
the appropriate Config method, which you then have to implement. If
you need a manual handler, you have to declare the option as manual in
job.yml and implement the handler yourself, though the automatically
generated code will declare it for you.

The build will fail until the new option is documented in
manual/cli.rst. To do that, create documentation for the option by
adding a ".. qpdf:option::" directive followed by a magic help comment
as described at the top of manual/cli.rst. Put this in the correct
help topic. Help topics roughly correspond with sections in that
chapter and are created using a special ".. help-topic" comment.
Follow the example of other options for style.

When done, the following should happen:

* qpdf --new-option should work as expected
* qpdf --help=--new-option should show the help from the comment in cli.rst
* qpdf --help=topic should list --new-option for the correct topic
* --new-option should appear in the manual
* --new-option should be in the command-line option index in the manual
* A Config method (in Config or one of the other Config classes in
  QPDFJob) should exist that corresponds to the command-line flag
* The job JSON file should have a new key in the schema corresponding
  to the new option


RELEASE PREPARATION

* Each year, update copyright notices. This will find all relevant
  places (assuming current copyright is from last year):

  git --no-pager grep -i -n -P "copyright.*$(expr $(date +%Y) - 1).*berkenbilt"

  Also update the copyright in these places:
  * debian package -- search for copyright.*berkenbilt in debian/copyright
  * qtest-driver, TestDriver.pm in qtest source

  Copyright last updated: 2022.

* Take a look at "External Libraries" in TODO to see if we need to
  make any changes. There is still some automation work left to do, so
  handling external-libs releases is still manual. See also
  README-maintainer in external-libs.

* Check for open fuzz crashes at https://oss-fuzz.com

* Check lgtm: https://lgtm.com/projects/g/qpdf/qpdf/?mode=list

* Check all open issues and pull requests in github and the
  sourceforge trackers. See ~/scripts/github-issues. Don't forget pull
  requests. Note: If the location for reporting issues changes, do a
  careful check of documentation and code to make sure any comments
  that include the issue creation URL are updated.

* Check `TODO` file to make sure all planned items for the release are
  done or retargeted.

* Check work `qpdf` project for private issues

* Run a spelling checker over the source code to catch errors in
  variable names, strings, and comments.

  make spell CLEAN=1

  This uses cspell. Install with `npm install -g cspell`. The output
  of cspell is suitable for use with `M-x grep` in emacs. Add
  exceptions to cSpell.json.

* If needed, run large file and image comparison tests. Configure
  options:

--enable-test-compare-images --with-large-file-test-path=/path

  For Windows, use a Windows style path, not an MSYS path for large files.

* Test with clang. Pass `CC=clang CXX=clang++` to `./configure`. (Done
  in CI).

* Test with newer version of gcc if available.

* Test 32-bit. Pass `CC=i686-linux-gnu-gcc CXX=i686-linux-gnu-g++` to
  `./configure`. (Done in CI.)

* Test build on a mac. (Done in CI.)

* Test with address sanitizer as described above. (Done in CI.)

* A small handful of additional files have been taken from autotools
  programs. These should probably be updated from time to time.

  * `config.guess`, `config.sub`, `ltmain.sh`, and the `m4` directory:
    these were created by running `libtoolize -c`. To update, run
    `libtoolize -f -c` or remove the files and rerun `libtoolize`. For
    `config.guess` and `config.sub`, search for "latest" in the files,
    and follow directions for updating them.

  * Other files copied as indicated:
    ```
    cp /usr/share/automake-1.16/install-sh .
    cp /usr/share/automake-1.16/mkinstalldirs .
    cp /usr/share/aclocal/pkg.m4 m4
    ```

  The entire contents of the `m4` directory came from `libtool.m4`. If
  we had some additional local parts, we could also add those to the
  `m4` directory. In order for this to work, it is necessary to run
  `aclocal -I m4` before running `autoheader` and `autoconf`. The
  `autogen.sh` script handles this.

* If any interfaces were added or changed, check C API to see whether
  changes are appropriate there as well. If necessary, review the
  casting policy in the manual, and ensure that integer types are
  properly handled with QIntC or the appropriate cast. Remember to
  ensure that any exceptions thrown by the library are caught and
  converted. See `trap_errors` in qpdf-c.cc.

* Double check versions and shared library details. They should
  already be up to date in the code.

  * Increment shared library version information as needed
    (`LT_CURRENT` in `configure.ac`)

  * Make sure version numbers are consistent in the following locations:
    * configure.ac
    * include/qpdf/DLL.h
    * manual/conf.py
    `make_dist` verifies this consistency.

  * Run ./autogen.sh

* Update release notes in manual. Look at diffs and ChangeLog.
  Update release date in `manual/release-notes.rst`.

* Add a release entry to ChangeLog: "x.y.z: release"

* Commit title: "Prepare x.y.z release"

* Performance test is included with binary compatibility steps. Even
  if releasing a new major release and not doing binary compatibility
  testing, do performance testing.

* Test for performance and binary compatibility:
  * Check out the last release
  * ./configure --enable-werror && make -j$(nproc)
  * Check out the current version
  * ./performance_check | tee -a /tmp/perf
  * ./configure --enable-werror && make -j$(nproc) build_libqpdf
  * Checkout the last release
  * make -k check NO_REBUILD=1 (some failures are normal -- looking
    for binary compatibility)
  * Check out the current version
  * make -j$(nproc)
  * ./performance_check | tee -a /tmp/perf

* Run pikepdf's test suite. Do this in a separate shell.

cd ...qpdf-source-tree...
export QPDF_SOURCE_TREE=$PWD
export LD_LIBRARY_PATH=$QPDF_SOURCE_TREE/libqpdf/build/.libs
cd /tmp/z
git clone git@github.com:pikepdf/pikepdf
virtualenv v
source v/bin/activate
cd pikepdf
pip3 install --upgrade pip
pip3 install '.[test]'
rehash
pip3 install .
pytest -n auto


CREATING A RELEASE

* Push to main. This will create an artifact called distribution
  which will contain all the distribution files. Download these,
  verify the checksums from the job output, rename to remove -ci from
  the names, and copy to the release archive area.

* Sign the source distribution:

version=x.y.z
gpg --detach-sign --armor qpdf-$version.tar.gz

* Build and test the debian package

* Add a calendar reminder to check the status of the debian package to
  make sure it is transitioning properly and to resolve any issues.

* Sign the releases. The release archive area should contain the
  Windows binaries, the AppImage, the source tarball, and the source
  tarball signature.

\rm -f *.sha256
files=(*)
sha256sum ${files[*]} >| qpdf-$version.sha256
gpg --clearsign --armor qpdf-$version.sha256
mv qpdf-$version.sha256.asc qpdf-$version.sha256
chmod 444 *
chmod 555 *.AppImage

* When creating releases on github and sourceforge, remember to copy
  `README-what-to-download.md` separately onto the download area if
  needed.

* Ensure that the main branch has been pushed to github. The
  rev-parse command below should show the same commit hash for all its
  arguments. Create and push a signed tag. This should be run with
  HEAD pointing to the tip of main.

git rev-parse qpdf/main @
git tag -s release-qpdf-$version @ -m"qpdf $version"
git push qpdf release-qpdf-$version

* Update documentation branches

git push qpdf @:$(echo $version | sed -E 's/\.[^\.]+$//')
git push qpdf @:stable

* If this is an x.y.0 release, visit
  https://readthedocs.org/projects/qpdf/versions/ (log in with
  github), and activate the latest major/minor version

* Create a github release after pushing the tag. `gcurl` is an alias
  that includes the auth token.

# Create release
GITHUB_TOKEN=$(qdata-show cred github-token)
function gcurl() { curl -H "Authorization: token $GITHUB_TOKEN" ${1+"$@"}; }
url=$(gcurl -s -XPOST https://api.github.com/repos/qpdf/qpdf/releases -d'{"tag_name": "release-qpdf-'$version'", "name": "qpdf '$version'", "draft": true}' | jq -r '.url')

# Get upload url
upload_url=$(gcurl -s $url | jq -r '.upload_url' | sed -E -e 's/\{.*\}//')
echo $upload_url

# Upload all the files. You can add a label attribute too, which
# overrides the name.
for i in *; do
  mime=$(file -b --mime-type $i)
  gcurl -H "Content-Type: $mime" --data-binary @$i "$upload_url?name=$i"
done

If needed, go onto github and make any manual updates such as
indicating a pre-release, adding release notes, etc.

Template for release notes:

```
This is qpdf version x.y.z. (Brief description)

For a full list of changes from previous releases, please see the [release notes](https://qpdf.readthedocs.io/en/stable/release-notes.html). See also [README-what-to-download](./README-what-to-download.md) for details about the available source and binary distributions.
```

# Publish release
gcurl -XPOST $url -d'{"draft": false}'

* Upload files to sourceforge.

rsync -vrlcO ./ jay_berkenbilt,qpdf@frs.sourceforge.net:/home/frs/project/q/qp/qpdf/qpdf/$version/

* On sourceforge, make the source package the default for all but
  Windows, and make the 32-bit mingw build the default for Windows.

* Publish a news item manually on sourceforge.

* Upload the debian package and Ubuntu ppa backports.

* Email the qpdf-announce list.


OTHER NOTES

For local iteration on the AppImage generation, it works to just
./build-scripts/build-appimage and get the resulting AppImage from
the distribution directory. You can also pass -e SKIP_TESTS=1
build-appimage, which passes it along to to docker, to skip the test
suite, which useful for rapid iteration.


GENERAL BUILD STUFF

QPDF uses autoconf and libtool but does not use automake. The only
files distributed with the qpdf source distribution that are not
controlled are `configure`, `libqpdf/qpdf/qpdf-config.h.in`,
`aclocal.m4`, and some documentation. See above for the steps required
to prepare a source distribution.

If building or editing documentation, configure with
`--enable-doc-maintenance`. This will ensure that all tools or files
required to validate and build documentation are available.

If you want to run `make maintainer-clean` or `make distclean` and you
haven't run `./configure`, you can pass `CLEAN=1` to make on the
command line to prevent it from complaining about configure not having
been run.

If you want to run checks without rerunning the build, pass
`NO_REBUILD=1` to make. This can be useful for special testing
scenarios such as validation of memory fixes or binary compatibility.


LOCAL WINDOWS TESTING PROCEDURE

This is what I do for routine testing on Windows.

From Windows, git clone from my Linux clone, and unzip
`external-libs`.

Look at `make_windows_releases`. Set up path the same way and run
whichever `./config-*` is appropriate for whichever compiler I need to
test with. Start one of the Visual Studio native compiler shells, and
from there, run one of the msys shells. The Visual Studio step is not
necessary if just building with mingw.


DOCS ON readthedocs.org

* Registered for an account at readthedocs.org with my github account
* Project page: https://readthedocs.org/projects/qpdf/
* Docs: https://qpdf.readthedocs.io/
* Admin -> Settings
  * Set project home page
  * Advanced
    * Show version warning
    * Default version: stable
  * Email Notifications: set email address for build failures

At this time, there is nothing in .github/workflows to support this.
It's all set up as an integration directly between github and
readthedocs.

The way readthedocs.org does stable and versions doesn't exactly work
for qpdf. My tagging convention is different from what they expect,
and I don't need versions for every point release. I have the
following branching strategy to support docs:

* x.y -- points to the latest x.y.z release
* stable -- points to the latest release

The release process includes updating the approach branches and
activating versions.