Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!brutus.cs.uiuc.edu!wuarchive!wugate!uunet!seismo!esosun!cogen!celerity!celit!ps
From: ps@celerity.UUCP (Patricia Shanahan)
Newsgroups: comp.std.c
Subject: Re: restrictive linkers (was: Reserved names in ANSI C)
Keywords: linkers, redux
Message-ID: <388@celit.UUCP>
Date: 25 Jul 89 17:51:38 GMT
References: <GEOFF.89Jun11231924@onyx.cs.warwick.ac.uk> <2619@yunexus.UUCP> <547@cybaswan.UUCP> <12711@bloom-beacon.MIT.EDU>
Sender: news@celerity
Reply-To: ps@celerity (Patricia Shanahan)
Organization: FPS Computing Inc., San Diego CA
Lines: 66

In article <12711@bloom-beacon.MIT.EDU> scs@adam.pika.mit.edu (Steve Summit) writes:
>Two minor points:
>
>In article <547@cybaswan.UUCP> iiit-sh@cybaswan.UUCP (Steve Hosgood) writes:
>>In article <2619@yunexus.UUCP>, davecb@yunexus.UUCP (David Collier-Brown) writes:
>>> 0) 6 characters is a lower limit of significance. You're allowed to use more.
>>Yeah, but though you may be able to use more, a 'strictly conforming' program
>>can't, otherwise it won't port to sites with 6-character linkers.
>
>Presumably the posters and most readers of this group understand
>it, but there are apparently many who interpret statements like
>the above to mean that external identifiers may not have more
>than six characters at all.  I'm always seeing subroutine
>packages that have incredibly strained, abbreviated identifier
>names, to keep them six characters long.  You're allowed to use
>extra characters for readability; the only problem is that if you
>say
>
>	int identifier1, identifier2;
>
>you may get
>
>	"identi: multiply defined."
>
>It should be perfectly legal to say
>
>	int identifier, anotheridentifier;

I strongly dislike systems that accept extra characters in
identifiers and ignore them. This feature creates a difference
between the program as the programmer reads it, and the program as
the compiler reads it. It can in fact move porting problems from
compile time detection to run time errors.

Suppose I have a large and complicated program that uses two
libraries, libA and libB. Suppose also that identifier1 is the name
of a function in libA and identifier2 is the name of a function in
libB. 

A short identifier system that treats ignored characters in
identifiers as an error will flag the use of more than six
characters in the function names during the library make, and will
also flag the use of the long names for external references during
the compile. The problem will be found and easily fixed before the first
successful make.

A short identifier system that simply ignores excess characters will
silently resolve references to both identifier1 and identifier2 to
whichever appears in the first library examined. If you are
lucky and have good tests, it will be found during program test.
Even after the existence of the bug is known because of a crash or
wrong result, it may still be difficult to fix. A programmer
reading the code will see the correct function being called. If the
system is sufficiently large, the programmer who is trying to fix a
failure in a  piece of code using identifier1 may not even be
conciously aware of the existence of identifier2. If they do a grep
for identifier1 it will not find identifier2.

This is especially serious if the significant identifier length is
scope dependent. The same type of problem can happen because the
scope of a function has been changed from static to external and it
has been moved to a library.
	Patricia Shanahan
	uucp : ucsd!celerity!ps
	arpa : ucsd!celerity!ps@nosc
	phone: (619) 271-9940