PATCH: read_poll Re: PATCH: Zpty cleanup (merge 13061 with 13116)

["Bart Schaefer" <schaefer@xxxxxxxxxxxxxxxxxxxxxxx>]

On Nov 8, 11:20am, Sven Wischnowsky wrote:
}
} Bart Schaefer wrote:
} 
} > However, that does not mean that the rest of the tests should be skipped
} > when select() returns 0.  A return of 0 means the select() timed out,
} > which (apparently) might happen under Cygwin even if there actually are
} > characters available to be read.  Peter/Andrej, is that the case?
} 
} I don't know about Cygwin, but that blocking read (line 1381) is
} exactly the test I wanted to avoid.

The read on line 1381 is *non-*blocking.  Since polltty is now always zero
when called from zpty, the setblock_fd() call is made before the read() is
attempted.  That was the patch I first committed which started me down the
road of merging in your changes.

On Nov 8,  1:42pm, Andrej Borsenkow wrote:
} Subject: RE: PATCH: Zpty cleanup (merge 13061 with 13116)
}
} I had not much time recently. Casual test shows, that zpty is currently
} completely broken on current Cygwin (1.1.5-4 as of this writing); I do not
} claim that it is zsh problem, but recent changes may have made it worse.

More on this in the other zpty thread ...

} About select - so far there seems to be no known problems under cygwin
} (that does not mean, there are no problems).

OK, if there are no known problems with select, then the test on line
1376 can change from `ret <= 0' to `ret < 0' -- and in that case it's
possible that we could split read_poll() into two cases, HAVE_SELECT v.
all the other code that's there.  I won't do that for now, though.

This still doesn't address my complaint that select() on my linux box
returns "ready for reading" when in fact the result of read() will be
an I/O error.  That makes no difference to `read -t' (for which the
read_poll() code was originally written) but it screws up the return
value of `zpty -r'.

On Nov 8,  5:03pm, Andrej Borsenkow wrote:
} Subject: read_poll RE: PATCH: Zpty cleanup (merge 13061 with 13116)
}
} read_poll tries to read from fd and stores the character it has read in
} readchar; that is cmd->read when called from checkptycmd. But this value does
} not seem to be used anywhere in read loop that looks like read_poll simply
} eats up this readahead char ...

You're correct.  The code Sven referenced at line 474 needs to be repeated
after line 484.

Index: Src/utils.c
===================================================================
@@ -1373,7 +1373,7 @@
 #endif
 #endif
 
-    if (ret <= 0) {
+    if (ret < 0) {
 	/*
 	 * Final attempt: set non-blocking read and try to read a character.
 	 * Praise Bill, this works under Cygwin (nothing else seems to).
Index: Src/Modules/zpty.c
===================================================================
@@ -482,6 +482,11 @@
 	    checkptycmd(cmd);
 	    if (cmd->fin)
 		break;
+	    if (cmd->read != -1) {
+		buf[++used] = (char) cmd->read;
+		seen = 1;
+		cmd->read = -1;
+	    }
 	}
 	if ((ret = read(cmd->fd, buf + used, 1)) == 1) {
 	    seen = 1;


-- 
Bart Schaefer                                 Brass Lantern Enterprises
http://www.well.com/user/barts              http://www.brasslantern.com

Zsh: http://www.zsh.org | PHPerl Project: http://phperl.sourceforge.net