Project

General

Profile

Actions

Bug #17012

closed

[versioning] generate correct development version numbers for all packages after a major release

Added by Ward Vandewege about 4 years ago. Updated over 3 years ago.

Status:
Resolved
Priority:
Normal
Assigned To:
Category:
-
Target version:
Story points:
-
Release relationship:
Auto

Description

The build/version-at-commit.sh script doesn't handle the scenario right after a major release well.

For code that hasn't changed since the major release, it generates version numbers without taking the major release tag into account.

For instance; as of commit c6c2f3518bc745eed95b5f5b81db5d17db4366ff on master, which is after the fork point of the 2.1-dev branch, these packages were generated for debian10:

arvados-api-server_2.1.0-1_amd64.deb
arvados-client_2.1.0~dev20201012021212-1_amd64.deb
arvados-controller_2.1.0~dev20201012021212-1_amd64.deb
arvados-dispatch-cloud_2.1.0~dev20201012021212-1_amd64.deb
arvados-docker-cleaner_2.2.0~dev20201015145956-1_amd64.deb
arvados-git-httpd_2.1.0~dev20201012021212-1_amd64.deb
arvados-health_2.1.0~dev20201012021212-1_amd64.deb
arvados-src_2.2.0~dev20201015145956-1_all.deb
arvados-sync-groups_2.2.0~dev20201013141632-1_amd64.deb
arvados-workbench_2.1.0~dev20201012021212-1_amd64.deb
arvados-ws_2.1.0~dev20201012021212-1_amd64.deb
crunch-dispatch-local_2.1.0~dev20201012021212-1_amd64.deb
crunch-dispatch-slurm_2.1.0~dev20201012021212-1_amd64.deb
crunch-run_2.1.0~dev20201012021212-1_amd64.deb
crunchstat_2.1.0~dev20201012021212-1_amd64.deb
keep-balance_2.1.0~dev20201012021212-1_amd64.deb
keep-block-check_2.1.0~dev20201012021212-1_amd64.deb
keep-exercise_2.1.0~dev20201012021212-1_amd64.deb
keepproxy_2.1.0~dev20201012021212-1_amd64.deb
keep-rsync_2.1.0~dev20201012021212-1_amd64.deb
keepstore_2.1.0~dev20201012021212-1_amd64.deb
keep-web_2.1.0~dev20201012021212-1_amd64.deb
libpam-arvados-go_2.1.0~dev20201012021212-1_amd64.deb
python3-arvados-cwl-runner_2.2.0~dev20201015145956-1_amd64.deb
python3-arvados-fuse_2.2.0~dev20201015145956-1_amd64.deb
python3-arvados-python-client_2.2.0~dev20201015145956-1_amd64.deb
python3-crunchstat-summary_2.2.0~dev20201015145956-1_amd64.deb

You can see that the python packages have a version number that takes the release into account (2.2.0~dev20201015145956), which is because there are some commits in master (post the 2.1-dev branch fork) that touch the Python part of the codebase.

For other parts of the codebase, a version string is generated that starts with 2.1.0~dev, which is not ideal.

For the API server package, it even generated the version string 2.1.0, because the calculated version for that package and its dependencies happens to be the commit that is tagged with the 2.1.0 version.


Subtasks 2 (0 open2 closed)

Task #17084: Review comment #6ResolvedWard Vandewege11/04/2020Actions
Task #17107: Python packages also depend on timestamp of build/version-at-commit.shResolvedWard Vandewege11/04/2020Actions
Actions #1

Updated by Ward Vandewege about 4 years ago

  • Status changed from New to In Progress
Actions #2

Updated by Ward Vandewege about 4 years ago

  • Description updated (diff)
Actions #3

Updated by Ward Vandewege about 4 years ago

  • Description updated (diff)
Actions #4

Updated by Ward Vandewege about 4 years ago

  • Description updated (diff)
Actions #5

Updated by Ward Vandewege about 4 years ago

  • Description updated (diff)
Actions #6

Updated by Ward Vandewege about 4 years ago

After some reflection, I think I've convinced myself that this behavior - counterintuitive as it is - is actually OK. The packages that are being generated do actually correspond with the source code from that version.

There is, however, a different bug. This all came up because a branch was merged that changed the package version numbers for dev packages from .dev to ~dev.

The version number generation code did not recognize that this change (obviously!) was a reason to generate new package version numbers. That's because that code never took the 'build' directory into account when calculating the in-tree dependencies of each of our pieces of software, the git hashes and timestamps of which are used to generate an appropriate version number.

So; the real fix here is to add the 'build' directory to the list of directories that is checked as part of the version number calculation.

This is easy for the Go and Rails packages, and I've pushed a fix along those lines in e94ddba9d544b173e5b56b41c6ac76ea0b072a26

For the Python packages, the version calculation is done in 'arvados_version.py', and that happens inside a virtualenv. It's more complicated to take the 'build' directory into account there, so I'm going to punt on that for now.

Actions #7

Updated by Ward Vandewege about 4 years ago

  • Target version changed from 2020-10-21 Sprint to 2020-11-04 Sprint
Actions #8

Updated by Peter Amstutz about 4 years ago

  • Target version changed from 2020-11-04 Sprint to 2020-11-18
Actions #9

Updated by Peter Amstutz about 4 years ago

Ward Vandewege wrote:

After some reflection, I think I've convinced myself that this behavior - counterintuitive as it is - is actually OK. The packages that are being generated do actually correspond with the source code from that version.

There is, however, a different bug. This all came up because a branch was merged that changed the package version numbers for dev packages from .dev to ~dev.

The version number generation code did not recognize that this change (obviously!) was a reason to generate new package version numbers. That's because that code never took the 'build' directory into account when calculating the in-tree dependencies of each of our pieces of software, the git hashes and timestamps of which are used to generate an appropriate version number.

So; the real fix here is to add the 'build' directory to the list of directories that is checked as part of the version number calculation.

This is easy for the Go and Rails packages, and I've pushed a fix along those lines in e94ddba9d544b173e5b56b41c6ac76ea0b072a26

For the Python packages, the version calculation is done in 'arvados_version.py', and that happens inside a virtualenv. It's more complicated to take the 'build' directory into account there, so I'm going to punt on that for now.

I don't think there's any problem taking the 'build' directory into account for versioning. The python code already does curdir+'/../../build/version-at-commit.sh' to get the version-at-commit script. The packages are always built from the source tree. It writes the version number to "_version.py" and during installation it just uses the version from "_version.py".

We could probably simplify arvados_version.py a little bit by separating the list of relative directories from the rest of the code. Unfortunately, consolidating it to one script that's symlinked everywhere is not so easy because Python packaging is weird about symlinks.

Actions #10

Updated by Peter Amstutz about 4 years ago

The other thing to be aware of is that any other place that computes the version needs to incorporate 'build' as well, e.g. run-library.sh calculate_python_sdk_cwl_package_versions()

Maybe that function could be changed so that it calls arvados_version.py

Actions #11

Updated by Ward Vandewege about 4 years ago

Peter Amstutz wrote:

We could probably simplify arvados_version.py a little bit by separating the list of relative directories from the rest of the code. Unfortunately, consolidating it to one script that's symlinked everywhere is not so easy because Python packaging is weird about symlinks.

Yeah. I've made a change along those lines.

The other thing to be aware of is that any other place that computes the version needs to incorporate 'build' as well, e.g. run-library.sh calculate_python_sdk_cwl_package_versions()

Yes, I've made the corresponding change. This function is only used in a few of our build scripts that make docker images.

Maybe that function could be changed so that it calls arvados_version.py

I didn't do that, it would have been much more complicated than this self-contained bash function.

Ready for a look in 185b8af696c553c0978f27e720c6924148af22fd on branch 17012-fix-python-version-calculation. Tests at developer-run-tests: #2181

Actions #12

Updated by Peter Amstutz about 4 years ago

The Python parts look great.

I don't think calculate_python_sdk_cwl_package_versions is quite correct.

cwl_runner_version is always going to be $max.

python_sdk_version is either python_sdk_version or build_version, whichever is later.

But we could also add something like this to arvados_version.py:

if __name__ == '__main__':
    print(get_version(SETUP_DIR, "arvados"))

Then calculate_python_sdk_cwl_package_versions() would just be:

calculate_python_sdk_cwl_package_versions() {
  python_sdk_version=$(cd sdk/python && python arvados_version.py)
  cwl_runner_version=$(cd sdk/cwl && python arvados_version.py)
}
Actions #13

Updated by Ward Vandewege about 4 years ago

Peter Amstutz wrote:

The Python parts look great.

I don't think calculate_python_sdk_cwl_package_versions is quite correct.

cwl_runner_version is always going to be $max.

It was actually correct - cwl_runner_version was never set to $max (which is a timestamp, but not clearly named as such) but rather to the latest of cwl/build/sdk.

But we could also add something like this to arvados_version.py:

[...]

Then calculate_python_sdk_cwl_package_versions() would just be:

[...]

OK, I've implemented that in 15d64feed20daa4a422ff9092615ac1e295d5ca2. I don't like that it's bash calling python calling bash, but the code is a bit simpler.

Actions #14

Updated by Peter Amstutz about 4 years ago

Ward Vandewege wrote:

Peter Amstutz wrote:

The Python parts look great.

I don't think calculate_python_sdk_cwl_package_versions is quite correct.

cwl_runner_version is always going to be $max.

It was actually correct - cwl_runner_version was never set to $max (which is a timestamp, but not clearly named as such) but rather to the latest of cwl/build/sdk.

But we could also add something like this to arvados_version.py:

[...]

Then calculate_python_sdk_cwl_package_versions() would just be:

[...]

OK, I've implemented that in 15d64feed20daa4a422ff9092615ac1e295d5ca2. I don't like that it's bash calling python calling bash, but the code is a bit simpler.

LGTM.

Actions #15

Updated by Ward Vandewege about 4 years ago

  • Status changed from In Progress to Resolved
Actions #16

Updated by Peter Amstutz over 3 years ago

  • Release set to 38
Actions

Also available in: Atom PDF