Re: Global alias mangles input in 5.0.8

[Peter Stephenson <p.w.stephenson@xxxxxxxxxxxx>]

On Wed, 01 Jul 2015 10:23:57 +0100
Peter Stephenson <p.stephenson@xxxxxxxxxxx> wrote:
> On Wed, 1 Jul 2015 00:48:23 -0400
> fcook <fcook@xxxxxxxxxxxxxx> wrote:
> > I'm running on x86-64 Arch Linux, and noticing what looks like a bug doing the following:
> > 
> > > alias -g '$'='4<<('
> > 
> > > cat /proc/self/fd/4 $ ls )
> > 
> > But under zsh 5.0.8, I get an error: "zsh: no such file or directory: s )".
> 
> Looks like bad effects are specific to "<(...)", since I seem to be getting
> sane results with "$(...)", but they're not specific to global aliases...
> 
> % alias foo='echo <('
> % foo ls )
> zsh: segmentation fault (core dumped)  ./zsh

I'm not sure why they don't happen with $(, but the reason is that we
have started to parse the complete (...) as a string so we can match
the correct closing parenthesis, and then we encounter the end of the
alias.  We backtrack because alias expansion occurs after the lexer,
while we're tracking the string at the level of input --- however, in
this case the alias was already being expanded when we found the "(",
so we backtrack too far.  The fix is not to do that.

Hope that's clear.

We didn't test this because it's an incredibly strange thing to do.

Luckily, there's absolutely no possibility of knock-on effects when
changing this code.  Ever.

Mikael...?

pws

diff --git a/Src/lex.c b/Src/lex.c
index baeed13..910773c 100644
--- a/Src/lex.c
+++ b/Src/lex.c
@@ -162,7 +162,7 @@ static int lex_add_raw;
 
 /* variables associated with the above */
 
-static char *tokstr_raw;
+static char *tokstr_raw, *lexbuf_ptr_start;
 static struct lexbufstate lexbuf_raw;
 
 /* text of punctuation tokens */
@@ -229,12 +229,13 @@ lex_context_save(struct lex_stack *ls, int toplevel)
     ls->lex_add_raw = lex_add_raw;
     ls->tokstr_raw = tokstr_raw;
     ls->lexbuf_raw = lexbuf_raw;
+    ls->lexbuf_ptr_start = lexbuf_ptr_start;
     ls->lexstop = lexstop;
     ls->toklineno = toklineno;
 
     tokstr = zshlextext = lexbuf.ptr = NULL;
     lexbuf.siz = 256;
-    tokstr_raw = lexbuf_raw.ptr = NULL;
+    tokstr_raw = lexbuf_raw.ptr = lexbuf_ptr_start = NULL;
     lexbuf_raw.siz = lexbuf_raw.len = lex_add_raw = 0;
 }
 
@@ -257,6 +258,7 @@ lex_context_restore(const struct lex_stack *ls, int toplevel)
     lex_add_raw = ls->lex_add_raw;
     tokstr_raw = ls->tokstr_raw;
     lexbuf_raw = ls->lexbuf_raw;
+    lexbuf_ptr_start = ls->lexbuf_ptr_start;
     lexstop = ls->lexstop;
     toklineno = ls->toklineno;
 }
@@ -1882,7 +1884,7 @@ zshlex_raw_add(int c)
 void
 zshlex_raw_back(void)
 {
-    if (!lex_add_raw)
+    if (!lex_add_raw || lexbuf_raw.ptr == lexbuf_ptr_start)
 	return;
     lexbuf_raw.ptr--;
     lexbuf_raw.len--;
@@ -1993,7 +1995,7 @@ skipcomm(void)
     cmdpop();
     return lexstop;
 #else
-    char *new_tokstr;
+    char *new_tokstr, *new_lexbuf_ptr_start;
     int new_lexstop, new_lex_add_raw;
     struct lexbufstate new_lexbuf;
 
@@ -2040,6 +2042,7 @@ skipcomm(void)
     }
     tokstr_raw = new_tokstr;
     lexbuf_raw = new_lexbuf;
+    lexbuf_ptr_start = lexbuf_raw.ptr;
     lex_add_raw = new_lex_add_raw;
     /*
      * Don't do any ZLE specials down here: they're only needed
@@ -2064,6 +2067,7 @@ skipcomm(void)
      */
     new_tokstr = tokstr_raw;
     new_lexbuf = lexbuf_raw;
+    new_lexbuf_ptr_start = lexbuf_ptr_start;
     /*
      * We're also going to propagate the lexical state:
      * if we couldn't parse the command substitution we
@@ -2079,6 +2083,7 @@ skipcomm(void)
 	 */
 	tokstr_raw = new_tokstr;
 	lexbuf_raw = new_lexbuf;
+	lexbuf_ptr_start = new_lexbuf_ptr_start;
     } else {
 	if (!new_lexstop) {
 	    /* Ignore the ')' added on input */
@@ -2093,6 +2098,7 @@ skipcomm(void)
 	tokstr = new_tokstr;
 	lexbuf = new_lexbuf;
 	lexstop = new_lexstop;
+	lexbuf_ptr_start = (char *)NULL;
 	hist_in_word(0);
     }
 
diff --git a/Src/zsh.h b/Src/zsh.h
index 183620f..d11d4fe 100644
--- a/Src/zsh.h
+++ b/Src/zsh.h
@@ -2801,6 +2801,7 @@ struct lex_stack {
     int lex_add_raw;
     char *tokstr_raw;
     struct lexbufstate lexbuf_raw;
+    char *lexbuf_ptr_start;
     int lexstop;
     zlong toklineno;
 };
diff --git a/Test/D03procsubst.ztst b/Test/D03procsubst.ztst
index c763f6e..07ec639 100644
--- a/Test/D03procsubst.ztst
+++ b/Test/D03procsubst.ztst
@@ -88,6 +88,7 @@
   print something=${:-=(echo 'C,D),(F,G)'}
 1: Graceful handling of bad substitution in enclosed context
 ?(eval):1: unterminated `=(...)'
+# '`
 
   () {
      print -n "first: "
@@ -115,3 +116,8 @@
 0:Process substitution as anonymous function argument
 >Execute a complicated order first
 >This line was brought to you by the letters F and D
+
+  alias foo='cat <('
+  eval 'foo echo this is bound to work)'
+0:backtacking within command string parsing with alias still pending
+>this is bound to work
diff --git a/Test/D08cmdsubst.ztst b/Test/D08cmdsubst.ztst
index a4c69a0..89e7259 100644
--- a/Test/D08cmdsubst.ztst
+++ b/Test/D08cmdsubst.ztst
@@ -148,3 +148,8 @@
  ) after
 0:Parsing of command substitution with ummatched parentheses: with frills
 >before start Universe began with u and ended with a crunch end after
+
+  alias foo='echo $('
+  eval 'foo echo this just works, OK\?)'
+0:backtracking within command string parsing with alias still pending
+>this just works, OK?