heredoc in command subst bug?

[Eric Blake]

I'm not sure whether this is a bug in POSIX or in bash, but I noticed the 
following with bash-3.2.39.

$ bash -c 'foo=$(cat <<EOF
hi
EOF)
echo $foo'

According to 
http://www.opengroup.org/onlinepubs/009695399/utilities/xcu_chap02.html#tag_02_0
7_04, heredoc parsing "continues until there is a line containing only the 
delimiter and a <newline>", and "EOF)" is not a line containing only the 
delimiter and a <newline>.  Therefore, per the rules in 
http://www.opengroup.org/onlinepubs/009695399/utilities/xcu_chap02.html#tag_02_0
6_03 that the closing ) is not located until a valid shell script has been seen 
as the command, this should be interpreted as an unterminated here-doc and an 
unterminated command-substitution, and cause a syntax error.  But bash treats 
the ) as the conclusion of the command-substitution, the EOF followed by end-of-
input as the conclusion of the heredoc, and prints out "hi" without error.

On the other hand, I suppose bash is doing something sensible (pdksh and zsh 
had the same behavior; only ash reported a syntax error), since this next 
example demonstrates that bash (and ash, zsh, and pdksh) consistently treats 
end-of-input as a valid heredoc terminator, even though no newline was present.

$ printf 'cat <<EOF
hi
EOF' > foo
$ chmod +x foo
$ bash ./foo
hi

A similar case occurs with ``:

$ bash -c 'foo=`cat <<EOF
hi
EOF`
echo $foo'

Here again, only ash behaves differently (but by printing "hi EOF" rather than 
complaining of a syntax error); but this time, POSIX is explicit that an 
unquoted ` in the body of a heredoc causes undefined behavior, and by my 
earlier arguments, 'EOF`' is still the body of the heredoc.  So this is not a 
valid POSIX test case, and bash neither needs to change behavior, nor needs to 
keep behavior the same if the $() case is changed.

Finally, bash has a definite bug, with no POSIX ambiguities:

$ bash -c 'cat <<EOF
hi'

Here, bash treats end-of-input as the heredoc delimiter, and prints "hi", even 
though it should be complaining of a syntax error since the heredoc (and thus 
the script) is unterminated.  pdksh behaves correctly and fails with a syntax 
error.

-- 
Eric Blake