Re: language design implications for variant records in a pascal-like language

From: "robin" <robin51@dodo.com.au>
Sent: Monday, 17 January 2011 11:55 AM


>> <compilers@is-not-my.name>


> Had spaces been significant, DO 10 I would have been parsed as three
> separate tokens. As it was, FORTRAN parsed it as the single token
> "DO10I", which was a legal identifier.
> [That's an egregious example, but I've written plenty of buggy code where
> I spelled a variable name in two ways. Not really a compiler issue, though,
> since it's easy enough to implement either way. -John]


The reason that many mis-spellings passed by unnoticed in FORTRAN was
that most compilers of the time produced only a compilation listing.


IBM PL/I compilers at that time produced not only a compilation
listing, but also an attribute listing (a list of identifiers,
attributes, and cross-references). That way it was easy to detect
mis-spelled identifiers.


Now, Fortran has not only free-form source, where blanks are significant --
which avoids constructs such as DO I = 1.10 becoming an assignment --
plus an IMPLICIT NONE statement which, when employed,
reports as errors any undeclared identifiers.


IBM's current PL/I compilers also have a compiler option that causes
undeclared identifiers to be classified as errors.

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