Programming Language Ranking History Data: Dataset Structure and Field Coverage

Every ranking series indexes a different proxy for popularity, and the proxy determines what your longitudinal study can claim. The three long-run sources in this industry split cleanly:

1. Search intent — PYPL recomputes monthly shares of Google Trends tutorial searches for 29 languages, so it tracks people learning to write a language.
2. Search-engine hits — TIOBE aggregates results across Google, Amazon, Wikipedia, Bing and more than 20 other engines, rating the top 50 languages each month.
3. Development activity — GitHut counts GitHub pull requests, pushes, stars and issues per language per quarter from Q2 2012 to Q1 2024, built on BigQuery githubarchive data.

None of them measures installed base, jobs or runtime minutes. That is why the same month can rank Python first everywhere while disagreeing about Rust versus C by several positions. Within the 24 primary systems-software datasets Datadory catalogs, these are the only three sources publishing a language-by-period panel rather than a single snapshot; everything else in this slice is either registry telemetry (npm, PyPI, crates.io), platform state (GitHub Octoverse) or survey opinion (Stack Overflow). Treat the three series as one triangulated instrument, not as competing measurements of the same quantity.

A reproducible merge takes six steps and fits in a notebook:

Step 5 is where sloppy studies break. A 2004-start PYPL curve next to a 2012-start GitHut curve invites readers to compare growth over unequal windows; rebase or truncate instead. And because PYPL smooths over roughly six months while GitHut reports raw quarterly counts, never read month-over-month divergence between the two as signal without checking lag.

Cite the series whose measurement matches your question, and say which one you used:

For toolchain or hiring questions, PYPL is usually the right headline because tutorial searches track people entering an ecosystem rather than machines rebuilding dependencies. For platform-strategy questions — which languages dominate open-source collaboration — GitHut's pull-request counts answer directly, with the caveat that the panel ends at Q1 2024. For academic citations needing respondent-level controls, the Stack Overflow microdata is the only row with individual-level observations. TIOBE earns its place as the longest continuous monthly brand, but quote its ratings as search-engine attention and expect to pay for machine-readable depth past the top 50.

Two of the three series have documented continuation paths inside this same catalog slice, which matters because GitHut froze at Q1 2024 and TIOBE's free surface resets each month.

GitHut is the clearest case: its README publishes the BigQuery SQL behind the site, filtering bot actors and dropping groups with fewer than 100 events from the githubarchive public datasets, so you can regenerate the quarterly panel for quarters after Q1 2024 using the identical query logic. The upstream GH Archive has mirrored GitHub public events into hourly gzip files and BigQuery since February 2011, and GitHub's REST API allows 5,000 requests per hour authenticated (60 unauthenticated) for spot checks. Because the filter thresholds are explicit, your extension stays comparable with the historical rows instead of drifting into a different definition of "a pull request."