On Sun, Jun 7, 2020 at 5:44 PM Ben Franksen
That said, he did write in one of his EWDs that beginners(!) should be tought programming with a language that does /not/ have an implementation available, in order to enourage thinking denotationally about programs rather than operationally, right from the start. The reason he preferred an imperative language for such a beginner's course (and also for communicating algorithms) is that recursion has a more complicated semantics (in terms of weakest preconditions) than his 'do' loop construct. This is important if, like Dijkstra did, you introduce every language construct by formally stating its denotational semantics, playing down the operational aspect as much as possible.
As a fun anecdote, Dijkstra's language referred to here, the Guarded Command Language, has actually seen several implementations despite his explicit intent for implementations never to exist: https://metacpan.org/pod/Commands::Guarded https://web.archive.org/web/20091214010836/http://www.dc.fi.udc.es/ai/~cabal... https://web.archive.org/web/20080622163307/http://gacela.labf.usb.ve/GacelaW... The last one there was actually used in the university I went to as the language for the first programming course for incoming CS students for the better part of a decade, until the department switched those courses to Python. Of note, also, is that 12 years passed from EWD1036 to his letter on using Haskell to teach introductory programming. I wonder if he ever wrote or shared his thoughts on whether this was merely a pragmatist's compromise, or a change of heart in light of Haskell's progress over those years: why did he not stick to GCL?