On Fri, 20 Feb 2004, Larry Hall wrote:
[on changing the error message when cp can stat but not open a file
due to .exe handling]
> How about a patch instead of a request? ;-)
Okay, attached. I've also attached a textutils patch to enact my earlier
request that "cat > foo" create foo in the whatever mode Cygwin would
usually want, rather than always using text mode.
> OK, I'm just kiddin' you. ;-) Think about a 'cp' that transparently handles
> '.exe' and a directory with 'foo' and 'foo.exe'. You need to be able to
> copy both. 'install' is a special purpose utility. Many projects that
> use it to install don't consider that executables may require extensions on
> other platforms. So they're written to just install 'foo'. That causes
> install failures. It makes sense to "relax" the rules here, since it
> helps more than it hurts. The reverse is true for 'cp' and other general
> purpose utilities.
As for being *able* to copy both foo and foo.exe, the special .exe
handling only occurs when *only* foo.exe exists. So you certainly can
copy both. As for install failures, I first started thinking about cp's
behavior due to a Makefile using cp rather than install.
That said, I do suppose the danger is less this way.
Hey, it might be interesting for managed mounts *really* to identify the
filenames "foo" and "foo.exe". (Maybe they already do?)
Robert
diff -Nrup fileutils-4.1-2.old/src/copy.c fileutils-4.1-2.new/src/copy.c
--- fileutils-4.1-2.old/src/copy.c 2001-06-25 23:42:46.000000000 +0100
+++ fileutils-4.1-2.new/src/copy.c 2004-02-21 17:34:32.742569600 +0000
@@ -197,6 +197,14 @@ copy_reg (const char *src_path, const ch
source_desc = open (src_path, O_RDONLY);
if (source_desc < 0)
{
+#ifndef __CYGWIN__
+ /* The following comment is incorrect.
+ The code that would have deferred the same-name check until
+ here is commented out in same_file_ok.
+ So in most environments, errno == ENOENT will never happen.
+ In Cygwin, this happens because of the different
+ handling of .exe in stat and open.
+ The error message for non-ENOENT errors is probably better. */
/* If SRC_PATH doesn't exist, then chances are good that the
user did something like this `cp --backup foo foo': and foo
existed to start with, but copy_internal renamed DST_PATH
@@ -205,6 +213,7 @@ copy_reg (const char *src_path, const ch
error (0, 0, _("%s and %s are the same file"),
quote_n (0, src_path), quote_n (1, dst_path));
else
+#endif
error (0, errno, _("cannot open %s for reading"), quote (src_path));
return -1;
diff -Nrup textutils-2.0.21.old/src/cat.c textutils-2.0.21.new/src/cat.c
--- textutils-2.0.21.old/src/cat.c 2001-12-01 17:29:26.000000000 +0000
+++ textutils-2.0.21.new/src/cat.c 2004-02-21 18:04:29.626361600 +0000
@@ -697,10 +697,13 @@ main (int argc, char **argv)
file_open_mode |= O_BINARY;
SET_BINARY (0);
+#ifndef __CYGWIN__
+ /* For Cygwin prefer default handling of text/binary on output */
/* Setting stdin to binary switches the console device to
raw I/O, which also affects stdout to console. Undo that. */
if (isatty (STDOUT_FILENO))
setmode (STDOUT_FILENO, O_TEXT);
+#endif
}
#endif
@@ -727,6 +730,9 @@ main (int argc, char **argv)
{
int tty_in = isatty (input_desc);
+#ifndef __CYGWIN__
+ /* For Cygwin prefer default handling of text/binary
+ on output */
/* If stdin is a terminal device, and it is the ONLY
input file (i.e. we didn't write anything to the
output yet), switch the output back to TEXT mode.
@@ -735,6 +741,7 @@ main (int argc, char **argv)
if (tty_in && optind <= argc)
setmode (STDOUT_FILENO, O_TEXT);
else
+#endif
{
SET_BINARY (input_desc);
# ifdef __DJGPP__
--
Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple
Problem reports: http://cygwin.com/problems.html
Documentation: http://cygwin.com/docs.html
FAQ: http://cygwin.com/faq/
