• Home
  • About
    • Anirban photo
    • About Me
    • Email
  • Blog Posts
    • Writings
    • Tags
  • Skill Set

data.table work diary: Mar 4 - 8

  • Mar 4, ‘24:

1) Did a first run of my GitHub Action to make a plot of all the documented tests with commit SHAs.

2) Made slides for my presentation tomorrow. The topic is GitHub Actions, and more specifically automated regression testing on pull requests. My slides can be found here.

3) Fixed the issue with commit IDs that Doris had.

  • Mar 5, ‘24:

1) Gave my presentation at SICCS, collected feedback, and discussed the changes I need to make to my GitHub Action.

2) Working on condensing different parts of my action.

3) Tried to fix the issues Doris is having with her example test cases - The first one uses atime::versions directly with a setup of the repository and that works, but the second version which includes the tests inside test.list and is passed onto atime::atime_pkg seems to throw a range of different errors for the same code being used. To investigate further.

  • Mar 6, ‘24:

1) Investigated ways to update a GitHub-bot comment and found different solutions but ended up on something much straightforward which existed for cml. Made that change to my action to have one single comment on a PR thread updated time and again instead of a new one for every commit.

2) Made an example to demonstrate my action (and atime too) running on another R package - Toby told me to test on bingsegRcpp; Forked it, made my changes, and got the action working: PR

  • Mar 7, ‘24:

1) Trying to run multiple jobs (matrix build/strategy) parallelly on different operating systems (Windows, Ubuntu, MacOS), and it’s currently failing for Windows (also for OS X but fixed the issues).

2) Investigated ways to benchmark different steps of a GitHub Action - Using Hyperfine, Recording UNIX timestamps before and after each step (reference to former one via IDs), from the REST API endpoints, webhooks (?)

3) Learned ways to branch out back in time to commits before and after a PR (reset with the PR’s parent and merge commit ID respectively), and investigated issues that reference #4491. (#5371 isn’t working for Doris locally, so skipping on that)

  • Mar 8, ‘24:

1) Made a new PR on my fork binsegRcpp to demonstrate a performance regression introduced manually.

2) Condensed my GitHub Action as Toby made atime available on CRAN again (installing atime from CRAN along with dependencies, thus removing the dependency installation step and the manual git2r setup).

3) Wrote code to benchmark the time taken by the atime step of my workflow (since that’s what we are concerned with and constitutes the majority of workflow runtime).

4) Tested the historical regression mentioned in #5424 as a result of merging PR #4491 (tried to replicate it ditto). Added the atime-based test case and got visible results in performance regression (PR).

5) Found that specification of a CRAN mirror was necessary for that old version (parent of #4491) of data.table. (#25)