.
Last update: 1997-05-20
9945-2-119
_____________________________________________________________________________
Topic: ed global commands
Relevant Sections: 4.20.7.3.7, 4.20.7.3.8, 4.20.7.3.25, 4.20.7.3.26
Defect Report:
-----------------------
From: sas@ironwood.cray.com (Scott Sutter)
Date: Thu, 27 Apr 1995 14:05:12 -0500 (CDT)
From: rcc@cray.com (Bob Clark)
Date: Apr 10, 1995
Dear Standards Board,
I would like to request an official interpretation from
WG15 concerning the following point in ISO/IEC 9945-2:1993 (POSIX.2).
The indicated sections of the ed utility description do not indicate
whether any given line in the buffer may be modified more than once
during a single global command invocation. (A global command refers to
any of the g, G, v, or V commands.)
Lines 3758-3760 of section 4.20.7.3.7 describe the basic premise of the
global commands:
In the global command, the first step shall be to mark every line
that matches the given BRE. Then, for every such line, the given
command list shall be executed with the current line number set to
the address of that line.
Lines 3781-3783 of section 4.20.7.3.8 indicate that the commands list
of the G command can modify any part of the buffer (this is also true
for the other global commands):
The commands read as part of the execution of the G command can
address and affect any lines in the buffer.
What comes into question is whether ed should allow a commands list to
modify any one line in the buffer on more than one iteration of a
global command.
Historically, the global commands do not execute the given command list
on any marked line which has already been modified (via a substitute or
move command in the command list), during the current global command
invocation.
For example, assume the following input file:
ab
bc
ac
ab
And the following ed command:
g/b/ 1,$s/$/./
Every ed command I have found gives the following output:
ab.
bc.
ac.
ab.
If the substitute command had been applied to the whole buffer (due to
the 1,$ address range) for each of the three /b/ matches found by the g
command, the resulting file would be:
ab...
bc...
ac...
ab...
This facet of the historical implementation prevents the extraordinary
consequences possible if lines may be modified more than once from
within any one global command.
The following problematic examples assume an ed command that allows
these multiple modifications. (The input file is the same as in the
example above.)
1) g/b/ .,$s/b/x/g
This substitution would fail after the first iteration since the
string to be substituted (/b/) has been removed.
2) g/b/ $s/a/aaaaaaaaaa/g
This substitution would cause the last line to grow in length at an
exponential rate. A similar case could split lines and make the
file length grow exponentially.
3) g/b/ +1m0
Here, the global search marked line 2 for later action, but on the
first iteration, it was moved to the top of the file. Should the
second iteration set dot to line 1 or line 4 before executing the
command list?
It is my opinion that the historical implementation of the global
commands is safer and is at least more straight forward than the
possible interpretation described above and should be clarified in
the relevant sections of POSIX.2.
Thank you for your attention to this matter.
Bob Clark
Interpretation response
------------------------
The standard states the behavior for ed global commands and
conforming implementations must conform to this. However, concerns have
been raised concerning historical practice which are being referred to
the sponsor.
Rationale
-------------
None.
Forwarded to Interpretations group: May 04 1995
Proposed resolution forwarded: Aug 11 1995
Finalized: Sept 12 1995