This document is produced by TC X3J14 as its clarification of questions raised about ANSI X3.215-1994, American National Standard for Information Systems - Programming Languages - Forth.

The questions covered herein were raised by query

Q0006, regarding Writing to Input Buffers.

There are four parts in this document:

  1. The original question as received.
  2. The TC's reply.
  3. The Letter Ballot issued by TC Chair.
  4. TC Chair's statement of ballot results.

Q0006 as received

Subject: Q0006 Request Recognized
To: X3J14@minerva.com
Date: Thu, 05 Oct 95 09:23:38 PDT  
The following query has been assigned number Q0005. - Greg Bailey, by direction 950913 0159Z
Request from TC member Jonah Thomas (Jet.Thomas@minerva.com)

Here's a question about the Standard:

When is it allowed to write into input buffers?

On the surface this is simple: "A program shall not write into the input buffer."

But what about this:

{
: CHANGE-TEXT S"  WORDS .S " S" ROT SWAP MOVE" EVALUATE ;

 .S .S .S .S .S SOURCE DROP CHANGE-TEXT .S
}
Our input buffer starts out with
 .S .S .S .S .S SOURCE DROP CHANGE-TEXT .S
When it executes CHANGE-TEXT it has ( ca ) the buffer address on the stack. CHANGE-TEXT first provides a string ( ca ca' u ) then another string ( ca ca' u ca" u") and then evaluates the second string. The original buffer is now not the input buffer, now the 2nd string is the read-only input buffer. It does ROT ( ca' u ca ) SWAP ( ca' ca u) MOVE and moves the first string into the original input buffer, which now reads
 WORDS .S .S .S .S SOURCE DROP CHANGE-TEXT .S
Then it reverts to the changed input buffer and does the final .S .

I say that even if this is legal, it's bad practice. If you want to alter an input buffer, better to copy it into your own private area and change it however you like and then do EVALUATE .

My question is whether this _is_ legal, and why or why not.

I originally interpreted a previous dpANS document as saying that the user input buffer, the one pointed to by the obsolescent TIB and #TIB , is read-only. That makes a different kind of sense to me. In some systems it could be a read-only pipe from some other process, or it could be some special hardware read-only buffer. It makes sense to me that the user input buffer should never be modified by a standard program.

So I can see 4 obvious choices, with perhaps others also available:

  1. The current input buffer is read-only; strings which are not the current input buffer may be written to unless they're read-only for some other reason.
  2. The current input buffer is read-only as is the user input buffer; any region of allotted memory is read/write except while it is the input buffer.
  3. Any region of memory that is EVALUATEd is forever after read-only.
  4. Any region of memory that is EVALUATEd is read-only until the evaluation is complete -- either by evaluation until the parse area is empty and return from EVALUATE or EXIT found in the evaluated string, or QUIT or ABORT performed before evaluation is finished, or if an ambiguous condition exists due to the code evaluated.
I don't like #3, since I sometimes want to evaluate text that's in block buffers or file buffers. #4 could take a lot of figuring to tell whether it's being violated or not. I like #1, but it seems to allow some atrocious practices in standard code. #2 has the further problem that there is no way for a program to tell whether an address is in the read-only user input buffer, except with SOURCE-ID or TIB , both of which are in an optional extension wordset.

Any ideas?

TC Reply to Q0006

From: Elizabeth Rather
Subject: Q0006R, Writing to Input Buffers, Official Response
To: X3J14 Technical Committee
Cc: lbarra@itic.nw.dc.us
Date: Mon, 19 Feb 96 14:09


Doc#:  X3J14/Q0006R
Reference Doc#:  X3.215.1994 ANS Forth
Date:  February 19, 1996
Title: Response to Request for Interpretation Q0006, Writing to Input Buffers

Q0006: When is it allowed to write into input buffers?

Request from TC member Jonah Thomas (Jet.Thomas@minerva.com)
Here's a question about the Standard: When is it allowed to write into input buffers?
When a standard program receives the address and length of the 'input buffer' from the word SOURCE it *must* treat this information as though it describes a read-only region of memory.

This is one of the conditions that must be met in order to compose code whose processing of 'input buffers' is independent of the physical and logical characteristics of the diverse 'input sources' described in the Standard.

It is logically possible for an application to alter the 'input buffer' supplied to EVALUATE while that 'input buffer' is being processed. This possibility exists as a special case because the application actually "owns" and has full control over the buffer in question. As owner of the buffer, and both producer and consumer of the data it contains, the application can alter this buffer and manipulate >IN with deterministic results, assuming that it knows the buffer to exist in physically writable memory.

The Standard does not describe this possibility because the implied coding methods constitute a special case which is only usable with EVALUATE and can lead to the development of source language syntax which is not processable from other 'input sources.'

Relevant text from the Standard:

6.1.1360 EVALUATE
... Make the string described ... both the 'input source' and 'input buffer' ...
6.1.2216 SOURCE
... c-addr is the address of, and u is the number of characters in, the 'input buffer'.
2.1 Definitions of Terms
input source:
The device, file, block, or other entity that supplies characters to refill the 'input buffer'.
input buffer:
A region of memory containing the sequence of characters from the input source that is currently accessible to a program.
3.3.3.5 Input buffers
The address, length, and content of the 'input buffer' may be transient. A program shall not write into the 'input buffer'. ... the 'input buffer' is either ... or a buffer specified by EVALUATE . ... An ambiguous condition exists if a program modifies the contents of the 'input buffer'."

Discussion of Technical Committee Intent

QUERY and TIB have been deprecated because while some systems have a discrete *place* called TIB that is exclusively used for storing the results of QUERY, other existing systems actually use TIB as a handle for the current line being interpreted from a file. While applications exist making either assumption, these applications can obviously not work properly on both sorts of systems in general.

Storing into 'input buffers' is disallowed because we permit input sources to nest indefinitely and it is not practical for systems that conserve resources to guarantee unique concurrent addressability of all nested input sources, nor is it practical to create separate save areas for all current input buffers just in case someone stored into one of them. The TC specifically intends that, when input is coming from refreshable sources, implementations may refresh their buffers on un-nesting to conserve resources, and that when logically possible implementations may use transient, shared buffers (as is common practice with LOAD on multiprogrammed systems.) Therefore, the results of storing into input buffers is stated as ambiguous, and may even be physically disallowed, as in the case of interpreting source from read only memory mapped files in some operating systems.

For similar reasons the address returned by SOURCE is transient, and there is specifically no guarantee that any 'input buffer' other than *the* (current) 'input buffer' is addressable at any time, nor that this address be valid after nesting or un-nesting. (Indeed, in classical multiprogrammed systems the address returned by SOURCE is no longer valid after using WORD .)

The TC expects all Systems to process buffers provided by EVALUATE in place. This is logically necessary, in our view, since there are no upper limits on the lengths of these buffers. Since it is semantically permissible to describe more than half of addressable memory in an EVALUATE string it is not in general *possible* to copy such a string elsewhere and address it consistently with the definition of SOURCE .

Systems are not allowed to alter the contents of 'input buffers' provided through EVALUATE . At the same time, applications are responsible for guaranteeing that the buffers they provide to EVALUATE are static in both address and content until the EVALUATE has completed.

EVALUATE is a special case since the input buffer is literally the area of memory provided to EVALUATE and is as such static. The application "owns" the memory it occupies, and the mechanism for messaging the interpreter via >IN (as well as the syntax of Forth) implies no prefetching or preprocessing of input buffers is necessary or appropriate.

Given these conditions, it *is* deterministic for an application to store (with great care) into EVALUATE buffers that it knows to be active, although such methods pertain exclusively to EVALUATE and certainly not to any other input stream source.

However, as the Standard is written, any program that does this is not a standard program.


It has been brought to the TC's attention that there exists at least one implementation which moves an EVALUATE string to a separate work area for processing. As a separate but related question we have been asked whether a system which does this, and which therefore fails John Hayes' test suite, is compliant. (The test in question fails if the buffer address and length returned by SOURCE occurring in an EVALUATE string differ from the address and length of the argument provided to EVALUATE.)

The system in question is definitely not compliant since both the letter of the standard and the intent of the TC are that EVALUATE strings be processed in place. The fact that John's test suite is able to detect the deviation confirms its visibility to application code, assuming that any such code ever makes such a comparison.

The ambiguity documented for storing into 'input buffers' serves to mitigate the seriousness of this noncompliance since any program which stores into an active EVALUATE string is environmentally dependent on the system's resolution of the ambiguity. Nevertheless, the Standard clearly states that the EVALUATE string *is* the 'input buffer' (by definition a region of memory), and that SOURCE returns the address and length of this region of memory. This is an unqualified promise to people writing standard programs, and any system which breaks that promise must document its nonstandard implementations of EVALUATE and SOURCE.

Letter Ballot

X3 Subgroup Letter Ballot
Authorized by X3 Procedures - Distributed by X3 Subgroup X3J14
Project: X3J14, ANS Forth
Doc#:  X3J14/LB017
Reference Doc#s:  X3J14/Q0006R, X3.215.1994 ANS Forth
Date:  February 19, 1996
Title:  Response to Request for Interpretation Q0006, Writing to Input Buffers
Ballot Period:  30 Days
Ballot Closes NOON DATE:  March 21, 1996
Respond to:  greg@minerva.com
        or:  Elizabeth D. Rather, Chair
             FORTH, Inc.
             111 N. Sepulveda Blvd.  Suite 300
             Manhattan Beach, CA  90266
             (310) 372-8493    FAX (310) 318-7130
             erather@forth.com

Statement:
    Document X3J14/Q0006R contains a proposed Response to Request for
    Interpretation Q0006.

Question:
    Do you agree that this response represents the intended interpretation of
    X3.215.1994 ANS Forth?


/------------------------  begin response area----------------------\
|
|  YES____ NO____ ABSTAIN____
|
|  Signature:  [not required for email ballots]
|  Name:
|  Organization:
|
|  Explanation (REQUIRED for NO or ABSTAIN votes):
|    
\------------------------  end response area  ----------------------/

INSTRUCTIONS:
Please return the entire letter ballot with your response _either_ by email
to greg@minerva.com _or_ by regular mail or fax or email to me at the above
address, before the closing date & time.

   If replying electronically PLEASE edit only within the response area
   indicated above, inserting any explanatory text in place of .
   Any changes made outside that area will likely be overlooked.

All TC members must vote.  Failure to vote in two consecutive ballots may
cause you to lose your voting rights in X3J14.

Thank you for your participation.

Elizabeth D. Rather, Chair, X3J14

Results of Letter Ballot

Letter ballot 17 closed at noon March 21 with the following results:

        Y  N  A NV
LB17:  12, 0, 1, 1

Abstention from John Hayes.