What language were the PL/I compilers written in?
I don't know about anyone else, but the VAX PL/1 front-end was bought by DEC from Freiburghouse (??SP??) in Framingham, MA. It was written in PL/1 on a Multics system. The Front-end was the same one that Pr1me used although Pr1me also bought their Fortran, which DEC did not. [FWIW: The DEC/Intel Fortran Front-End was written in Pascal -- still is last time I talked to the compiler folks].
I do not know what the Freiburghouse folks used for a compiler-compiler (Steve or Doug might ), but >>I think<< it might not have used one. Culter famously led the new backend for it and had to shuttle tapes from MIT to ZKO in Nashua during the development. The backend was written in a combination of PL/1, BLISS32 and Assembler. Once the compiler could self host, everything moved to ZKO.
That compiler originally targeted VMS, but was moved to Unix/VAX at one point as someone else pointed out.
When the new GEM compilers were about 10-15 years later, I was under the impressions that the original Freiburghouse/Culter hacked front-end was reworked to use the GEM backend system, as GEM used BLISS, and C for the runtimes and a small amount of Assembler as needed for each ISA [And I believe it continues to be the same from VSI folks today]. GEM based PL/1 was released on Alpha when I was still at DEC, and I believe that it was released for Itanium a few years later [by Intel under contract to Compaq/HP]. VSI has built a GEM based Intel*64 and is releasing/has released VMS for same using it; I would suspect they moved PL/1 over also [Their target customer is the traditional DEC VMS customer that still has active applications and wants to run them on modern HW]. I'll have to ask one of my former coworkers, who at one point was and I still think is, the main compiler guy at VSI/resident GEM expert.
Wikipedia claims that IBM is still developing a PL/I compiler, which I suppose I have no reason to disbelieve, but I'm very curious as to who is using it and for what purpose.
As best I can tell, commercial sites still use it for traditional code, just like Cobol. It's interesting, Intel does neither but we spend a ton of money on Fortran because so much development (both old and new) in the scientific community requires it. I answered why elsewhere in more detail: Where is Fortran used these days and Is Fortran still alive
My >>guess<< is that PL/1 is suffering the same fate as Cobol, and fading because the apps are being/have been slowly rewritten from custom code to using COTS solutions from folks like Oracle, SAS, BAAN and the like. Not so for Fortran and the reason is that the math has not changed. The core of these codes is the same was it was in the 1960s/70s when they were written. A friend of mine used to be the Chief Metallurgist for the US Gov at NIST and as Dr. Fek put it so well: "I have over 60 years worth of data that we have classified and we understand what it is telling us. If you magically gave me new code to do the same thing as what we do with our processes that we have developed over the years, I would have to reclassify all that data. It's just not economically interesting." I personally equate it to the QWERTY keyboard. Just not going to change. i.e. "Simple economics always beats sophisticated architecture."