[sr #110654] On AT&T UnixPC (3b1) autoconf script fails to properly replace $LINENO variable with line number

[Alain Knaff]

URL:
  <https://savannah.gnu.org/support/?110654>

                 Summary: On AT&T UnixPC (3b1) autoconf script fails to
properly replace $LINENO variable with line number
                 Project: Autoconf
            Submitted by: alainknaff
            Submitted on: Sun 15 May 2022 01:07:40 PM CEST
                Category: None
                Priority: 5 - Normal
                Severity: 3 - Normal
                  Status: None
                 Privacy: Public
             Assigned to: None
        Originator Email: 
             Open/Closed: Open
         Discussion Lock: Any
        Operating System: None

    _______________________________________________________

Details:

Hi,

Autoconfigure scripts generated by GNU autoconf pass themselves through a pair
sed script to substitute $LINENO variable with the actual line number if the
shell itself does not support a $LINENO variable.

The first sed script adds a line with just the line number after each line
containing $LINENO.
The second script then substitutes $LINENO with the contents of next line (the
line number placed by first script)

This second script fails on AT&T UnixPC (3b1) and leaves some of the line
numbers in the file on their line on their own.


This is due to a different behavior of sed's "t" command.
The "t" command branches to the given label if there has been a successful
s/// substitution since the last t command.
On most other platforms, "t" only considers substitutions within current
"line". However AT&T UnixPC's sed also considers substitutions on previous
lines.

This breaks the second script if there are 2 lines with LINENO variable that
have *exactly* one intervening other line between them.

Example:
abc $LINENO
def
fgh $LINENO

This is transformed by the first sed script into
abc $LINENO
1
def
fgh $LINENO
3

The second script first processes the first 2 lines into "abc 1"
However, the removal of the "1" from its own line (s/-\n.*//) is counted as a
successful substitution, and is remembered (by UnixPC's sed) for the *next*
line def (other sed implementations would clear this on passing to next line).
As the sed script now thinks that def contains a $LINENO variable, it gobbles
up the next line "fgh $LINENO" using N command, substitutes $LINENO with
following line (i.e. with nothing, as it hasn't read the 3 into pattern space
yet => eventually, it moves on to the next line (the unprocessed 3), and
leaves the line number as is in the output, on a line by its own, where this
number ends up getting executed as a command, which crashes the script.

Fix: this can be addressed by putting an empty branch + label at the end of
the second sed script (t end \n :end):

...
    sed '
      s/[$]LINENO.*/&-/
      t lineno
      b
      :lineno
      N
      :loop
      s/[$]LINENO\([^'$as_cr_alnum'_].*\n\)\(.*\)/\2\1\2/
      t loop
      s/-\n.*//
      t end
      :end
    ' >$as_me.lineno &&
...

in /usr/share/autoconf/m4sugar/m4sh.m4

N.B. For testing, an AT&T UnixPC emulator can be found at
https://github.com/philpem/freebee






    _______________________________________________________

Reply to this item at:

  <https://savannah.gnu.org/support/?110654>

_______________________________________________
  Message sent via Savannah
  https://savannah.gnu.org/