Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!wasatch!cs.utexas.edu!sm.unisys.com!ism780c!news
From: news@ism780c.isc.com (News system)
Newsgroups: comp.arch
Subject: Re: String lengths
Message-ID: <22036@ism780c.isc.com>
Date: 8 Feb 89 21:03:41 GMT
References: <8882@alice.UUCP>
Reply-To: marv@ism780.UUCP (Marvin Rubenstein)
Organization: Interactive Systems Corp., Santa Monica CA
Lines: 45

In article <8882@alice.UUCP> dmr@alice.UUCP writes:
>                                        More nearly correct
>memory--which applies only to 20-year-old implementations
>on the IBM 7090 and GE 635/645--says that strings existed
>in two forms: a one-byte count followed by the characters
>of the string, and an expanded form in which the count
>and the characters were blown out into words.
>
>In these implementations, bytes were 9 bits, and so the compact
>representation limited strings to length 511.  On the
>other hand, the blown-out form occupied lots of space and
>wasn't suitable for transferring directly to text files.
>Originally, I believe, there were no operators for accessing
>individual characters in a compact string, though they were
>added later.

I cannot speak about the GE 635, But I can speak with some authority about
the IBM 7090.  I have before me both the 7090 reference manual and the micro
code that I wrote to emulate the 7090 on the Standard IC4000.

The 7090 provided no instructions for accessing individual characters.  The
notion of characters existed only at the I/O interface where 6 characters
(in 6 bit BCD) could be read into a single 36 bit word.  The minimum transfer
was two words.  I am not aware of any languages made available by IBM that
provided a string data type.  In fact the limit of 6 character identifiers in
FORTRAN was due to the fact that 6 characters could be manipulated as a word
on the 7090 (and on the 704 the original "FORTRAN" machine).

IBM did produce a 7040 machine (similar to but not an extension of the 7090).
This machine provided character load and store.  But the use of the facility
was limited by the fact that a character address could not be put into an
index register.

I implemented an extended version of 7090 called the EX02 by Standard
Computer that did support strings.  Because the words had room in them for
two pointers and a character, I represented a string as a linked list with
one character per word.  The ends of the string was marked by null pointers.
There were instructions to traverse the linked list and to convert from a
packed array of characters to a string and back again.  The language that
took advantage of these strings was called IMPLAN (implementation language).
An interesting feature of the language was that an expression like i*s, where
i was an integer and s was a string, meant the first i charcters of s.  And
s*i meant the last i characters of the string.

   Marv Rubinstein