In several subfields of CS (e.g. PL, HCI, graphics) there's a feeling that good systems research is too hard to publish.

In programming languages, SIGPLAN actually has an award for the most impactful systems. So I wondered — how "publishable" were papers on those systems?

Of the 12 winners, only 3 — WebAssembly, Pin, and LLVM — had a canonical systems paper published in a SIGPLAN venue (conference or journal).

Another four — Racket, Z3, Jikes, and GHC — had a systems paper published in another major publication.

I'd count these as "misses" for SIGPLAN, as they were all published in venues that are ranked below the top SIGPLAN venues (PLDI, POPL, TOPLAS, etc.)

(Of course, my assumption here is that the authors would have preferred to publish in one of the top SIGPLAN venues if they could have. If that's wrong, I'm happy to be corrected here.)

Of the remaining 5 —

• The CompCert and Scala papers weren't published in major venues (not indexed by dblp)
• Coq doesn't have a canonical publication that I'm aware of
• Neither do V8 and GCC; they also don't come from academia, so maybe the authors didn't care to publish

So, ignoring V8 & GCC…

Of 10 projects that won a SIGPLAN award for "significant impact on PL research, implementations & tools"

only 3 had a systems paper in a SIGPLAN venue.

That seems low to me, and suggests that it is indeed too hard to publish systems work in those venues

Caveats and fine print —

The full list of papers is here: docs.google.com/spreadsheets/d…. I used citation counts and abstracts to judge what the "canonical systems paper" was for each project. If you think I got it wrong, let me know!

A correction —

a Jikes system paper (it was called Jalapeño then) was published at OOPSLA '99.

So, 4 of 10 had a systems paper in a SIGPLAN venue.

twitter.com/JAldrichPL/sta…