x=<< functionality cannot be used in functions

[Thorsten Glaser]

Several bugs with this construct:

x=<<EOF
foo
EOF

x+=<<
foo
<<

1. It cannot be used in functions: assignment is done at parse time
2. It uses ATEMP of parse time, which is gone at runtime in functions
   ⇒ hides first bug unless exec.c alloc(z * 2 + n, ATEMP) → APERM
3. Size calculations not fully checked, one too big
   ⇒ (z-1) * 2 + 1 /* EOS */ + n

Ad 1:

(sid-amd64)tglase@tglase:~/b $ ./mksh ~/x
a
abla5424 bar
abla5424 bar bla5424 bar
function bar {
        x+=\b\l\a\5\4\2\4\ \b\a\r\

}
(sid-amd64)tglase@tglase:~/b $ cat ~/x
function bar {
x+=<<
bla$RANDOM bar
<<
}
x=a
echo $x
bar
echo $x
bar
echo $x
typeset -f bar

Substitution happens at first use:

(sid-amd64)tglase@tglase:~/b $ ./mksh ~/x2
function bar {
        x=<<
bla$RANDOM bar
<<

}
function bar {
        x=\b\l\a\1\2\2\8\1\ \b\a\r\

}
(sid-amd64)tglase@tglase:~/b $ /bin/mksh ~/x2
function bar {
        x=<<
bla$RANDOM bar
<<

}
function bar {
        x=\b\l\a\2\6\7\7\9\ \b\a\r\

}
(sid-amd64)tglase@tglase:~/b $ cat ~/x2
function bar {
x=<<
bla$RANDOM bar
<<
}
typeset -f bar
bar
typeset -f bar

So this is even worse than thought, the ATEMP context is the one from the first use of the function…

This was introduced in R40, when the x=<< syntax was added.

[Thorsten Glaser (mirabilos)]

This additionally requires a reparse hack (see lex.c for IONDELIM) because the tree.c printing code appends a trailing newline to all constructs unconditionally; we can remove that when the latter is no longer true (just documenting).

[Thorsten Glaser (mirabilos)]

fix in commitid 100576F1D4100EFF065