While I am certain we were not the first to use the term, I'm pretty
certain that we were using "middle end" to describe parts of the TSI
compilers (the same PL/I compiler DEC bought and a series of other
ones written by various companies (including in house) that used many
of the same components that were neither completely language nor
completely target specific. This would have been around 1985 or so.

The "front ends" consisted of the lexers and parsers for the languages.

The "back end" was essentially just the code generator (and peephole
optimizer once we implemented one).

The "middle end" included the symbol table, the intermediate language,
the storage allocation routines (i.e. how stack frames and static
memory areas were laid out since we used the same layout for all the
architectures), and the global optimizer. And the phases front,
middle, and back were sequentially ordered in terms of how they were
executed. I.e. The front end ran first, then the middle end, and
finally the back end, although the symbol table and intermediate
language both spanned the entire compilation process. The compiler
option processing was also included in the middle end even though it
preceded the front end.

Now, we grouped the latter two into the "common envelope" (although we
originally called it just the "common backend" and thought of the
middle end as part of the back end). The common envelope team grew to
be almost as many people as the sum of the front end teams put
together and had a full-time manager running the team.

Now, some parts of both the middle and back ends had language specific
hooks in them (mostly for FORTRAN or COBOL, but a few for Pascal and
C). The C compiler was short-lived though and a completely new one
was written.

--
******************************************************************************
Chris Clark email: christopher.f.clark@compiler-resources.com
Compiler Resources, Inc. Web Site: http://world.std.com/~compres
23 Bailey Rd voice: (508) 435-5016
Berlin, MA 01503 USA twitter: @intel_chris
------------------------------------------------------------------------------
[FWIW, the term first appeared in comp.compilers in 1993. -John]

Chris,

> While I am certain we were not the first to use the term, I'm pretty
> certain that we were using "middle end" to describe parts of the TSI
> compilers (the same PL/I compiler DEC bought and a series of other
> ones written by various companies (including in house) that used many
> of the same components that were neither completely language nor
> completely target specific. This would have been around 1985 or so.

The term middle-end was being used at Intermetrics when I was there
in 1981.

The people there were strongly influenced by the PQCC work, but the
project I worked on had just three stages, not the umpteen stages used
by PQCC. Perhaps it came from a project that an Intermetrics person
worked on.

Derek Jones recollection of its usage at Intermetrics with a possible
connection to the PQCC work done at CMU sounds quite plausible. Many
of us were influenced by that work. I would have been just leaving
SofTech at that time and joining Pr1me Computer (where my reference to
the term came from. By the way, I did not work at TSI, just to be
clear, just maintained their compiler backends while at Pr1me.) I
don't recall either way whether the term was used at SofTech. I was
simply too junior at that time.

--
******************************************************************************.
Chris Clark email: christopher.f.clark@compiler-resources.com
Compiler Resources, Inc. Web Site: http://world.std.com/~compres
23 Bailey Rd voice: (508) 435-5016
Berlin, MA 01503 USA twitter: @intel_chris
------------------------------------------------------------------------------

Chris,

> Derek Jones recollection of its usage at Intermetrics with a possible
> connection to the PQCC work done at CMU sounds quite plausible. Many

The CHILL compiler team was huge, by compiler standards.
We actually had front-end, middle-end, and back-end teams.
I have not since worked on a compiler project with this exact
explicit breakdown (which was probably driven by the division
of multi-company labor).

I know there were earlier compilers at Intermetrics that had a
middle-end concept. Tony Flanders knows a lot of history and might
be able to say where the term originated:
http://www.whysheep.com/i2/daf-history2.html