Re: Compiler stress tests?

G. Herrmannsfeldt wrote:
> I used to hear, int the days of punched cards, that some compiler
> testers would get cards from the recycle bins and feed them to the
> compiler. Just random collections of possibly error containing
> statements, unrelated to each other.


On a related note, a suite of Algol68 compiler killers had a pair of
programs: the first was a legal and reasonable program that broke up
Algol68 programs lexically and spat out the lexemes in random order.
The second program was this random output. Generally speaking, this
played havoc with any error recovery since your typically error
recovery tries to resynchronise on something sensible like ";" or
"END".


I guess the underlying reason it was so cruel to compilers is that the
random program is almost entirely, but not quite, unlike an Algol68
program.


[The possibility of generating legal but highly complex programs has
been mentioned already -- this gives something with is almost entirely,
but not quite, unlike a working program.]


--
John Haxby
jch@pwd.hp.com http://www.openmail.external.hp.com/
--

Return to the comp.compilers page.
Search the comp.compilers archives again.