Explanation of distprompter

The Section distprompter Edits dist Drafts gives an overview of this script. Here is the distprompter script (you might want to open it in a separate browser window). To install it, see the Section Programs in This Book's Archive.

The main part of distprompter is the while loop. It reads the template draft file (that dist built; the Section Making the Draft from the Template File explains how). The draft filename is in the first command-line parameter, $1. The while loop writes the edited header into the $header file. The loop is shown below. The line numbers like 31> aren't part of the file; they're just for reference.

30> # read COMMAND AT TOP OF LOOP GETS STDIN (FD 0), SO SAVE FD 0 NOW:
31> exec 4<&0   # SAVE ORIGINAL STDIN (USUALLY TTY) AS FD 4
32> while read label line
33> do
34>     case "$label" in
35>     [Rr]esent-?*:)
36>         case "$line" in
37>         ?*) # SHOW LINE ON SCREEN AND PUT INTO HEADER FILE:
38>             echo "$label $line"
39>             echo "$label $line" 1>&3
40>             ;;
41>         *)  # FILL IT IN OURSELVES:
42>             $echo "$label $nnl"
43>             exec 5<&0   # SAVE DRAFT FILE FD; DO NOT CLOSE!
44>             exec 0<&4   # RESTORE ORIGINAL STDIN
45>             read ans
46>             exec 0<&5   # RECONNECT DRAFT FILE TO STDIN
47>             case "$ans" in
48>             "") ;;  # EMPTY; DO NOTHING
49>             *)  echo "$label $ans" 1>&3 ;;
50>             esac
51>             ;;
52>         esac
53>         ;;
54>     ""|---*) # END OF HEADER
55>         echo "-------" 1>&3
56>         break   # PROBABLY NOT NEEDED...
57>         ;;
58>     *)  echo "$myname: illegal header component
59>         '$label $line'" 1>&2
60>         break
61>         ;;
62>     esac
63> done <$1 2>$err 3>$header

The Table below lists the file descriptor numbers used in operators like 4<&0.

Table: UNIX File Descriptor Numbers

Number   Default Use

0        Standard input

1        Standard output

2        Standard error

3-9      Not assigned*

Outside the loop, the standard input of the commands is (by default) from your keyboard. But the loop is different because of the redirection at the done line (line 63 -- see below). Every command in the loop reads its standard input from the template draft file, not from your terminal. It "takes over" standard input during the loop. So, the read command at the top of the loop (line 32) reads the template draft file. But that means the read command inside the loop, on line 45, couldn't read from your terminal.

The answer is to connect any other files you want to read to their own file descriptors. Line 31 does that. Inside the loop, instead of reading its standard input, the read on line 45 reads file descriptor 4.

Before reconnecting to the saved file descriptor 4, there's one more thing the script needs to do. Line 43 saves the open draft file template as file descriptor 5. Because we don't close the file here (with exec 5<&-), the next read command at the top of the loop (line 32) will read the next line of the file. If we'd simply closed the template file and reopened it, the next read on line 32 would read the first line of the file instead of the correct line.

Next, line 44 changes file descriptor 0 to point to the original standard input before the loop -- it was saved as file descriptor 4 in line 31. Now, the read ans command on line 45 can read the original standard input and get your answer to the prompt. Line 46 rearranges descriptors for the rest of the loop -- it reconnects the draft template file from file descriptor 5 back to file descriptor 0. So, the next read at the top of the loop will read the template file.

Some shells can solve the problem on line 45 directly. They let you change what file descriptor the read command reads, instead of the default (file descriptor 0, the standard input). For instance, to read the value of the ans variable from file descriptor 4, line 45 could look like one of these:

read ans 0<&4
read -u4 ans

That's why distprompter was written this way: the kind of redirection around line 45 works on all Bourne shells.

Now, about the loop: three file descriptors (fds) are redirected at line 63; these affect the input and output of every command inside the loop:

fd 0

The standard input; taken from the draft template file. Each read command on line 32 will read the next line from the template file.

Any other command inside the loop that reads its standard input would read from this template file, too. That would be bad here, of course. See the explanation of lines 43-47, above.

fd 2

The standard error; goes to an error-collecting file.

fd 3

Written to the new message header file. Anything that is written to file descriptor 3 during the loop will be added to the new header.

The read in line 32 reads two shell variables. The first variable, label, always contains the first word on the line -- usually a field label like Resent-Fcc:. If the draft file has any other text (like the word outbox), that goes into the line variable.

The body of the loop is a big case structure. It tests the label and branches:

• If the label starts with Resent-, lines 35-53 test to see if the rest of the field had any text.
  • A complete field is echoed to the terminal (via standard output) in line 38 and to the new header file (by file descriptor 3) in line 39.
  • If the draft field was empty, line 42 echos it without a newline. This leaves the cursor to the right of the colon, ready for the user's response. The $echo and $nnl set at one of the lines 12-14 makes this work portably.

    A simple read command won't read what you type at the prompt because the loop's standard input comes from the draft file. As you read at the start of this section, some shells' built-in read command will not let you redirect their standard input to another file descriptor. A workaround that should take care of all Bourne shells starts at line 43. It's explained in detail near the start of this section.

    The case in lines 47-50 tests the response. The field is only printed to the new header if the user typed an answer.

• The case in line 54 matches an empty line or a line that starts with at least three dashes. This is the end of the header. The script echos its own row of dashes to the draft file. Then it breaks the loop -- that probably isn't necessary because the next read in line 32 should return a nonzero status when it reads the end of the template file.
• Line 58 matches anything else in the header. Because MH is strict about the fields in the dist header, this test does some early enforcement of the rules and avoids those problems when you try to send the message.

[Table of Contents] [Index] [Previous: Explanation of autoinc] [Next: Explanation of edprofile]

This file is from the third edition of the book MH & xmh: Email for Users & Programmers, ISBN 1-56592-093-7, by Jerry Peek. Copyright © 1991, 1992, 1995 by O'Reilly & Associates, Inc. This file is freely-available; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation. For more information, see the file copying.htm.