r63214 MediaWiki - Code Review archive

The reference quoted in comment 11 on the bug report (http://technet.microsoft.com/en-us/library/cc723564.aspx) states that special characters such as ^ must either be escaped with a ^, or enclosed in double quotes. The reference, and experimentation, suggests that doing both (as this code does) is incorrect:

C:\>echo ^^
^

C:\>echo "^^"
"^^"

C:\>echo "^"
"^"

I'm reverting. If you still think you're right, then you should respond on the bug report and document your testing procedure fully.

Do not use echo to test the Windows shell. It seems to be a builtin sometimes using different rules (eg. the quotes are printed as literals).

I tested it by compiling a program that iterated argv printing it. The results are in bug 13518#c9.

Math.php calls several times escapeshellarg() which, under Win32 will be surrounding with double quotes.

Thus we end up doing a call like wfEscapeShellArg('foo "bar" "baz"'); which outputs "foo \"bar\" \"baz\"". The \" escape the quotes but oddly the carets are treated as "not inside quotes", regardless of the closing quote being escaped with a backslash.

So relying only on the external double quotes is wrong and always doubling the carets isn't right either.

Please give the source code and compiler details of the program you used to test it.

You're probably aware that unlike UNIX, Windows passes command line parameters to applications as a single string, which the called program is then required to parse. Typically this is done by the Visual C++ run-time library, which splits the string provided and populates the argv and argc variables. The question then is: at what stage are double-carets converted to single-carets, and what is the relevant algorithm?

If the algorithm is something different to the documented behaviour for CMD.EXE, then we probably need to do a disassembly of the relevant VC++ code.

Yes, I know. I have been looking into it with more detail. The caret is handled by cmd.exe, while the quote escaping is at the program level. The problem is that in "Foo\"Bar", the content enclosed in double quotes is Foo\.

Program to output arguments:

#include <string.h>
#include <windows.h>

int main(int argc, char** argv) {
	int i;
	LPWSTR* argv2;
	
	puts( "GetCommandLine:" );
	puts( GetCommandLineA() );
	
	puts( "\n__getmainargs:" );
	do {
		printf( "<%s> ", *argv );
	} while( *++argv );

	puts( "\n\nCommandLineToArgvW:" );
	argv2 = CommandLineToArgvW( GetCommandLineW(), &argc );
	for ( i = 0; i < argc; i++) {
		wprintf( L"<%ls> ", argv2[i] );
	}

	return 0;
}

Testcase showing the problem:

<?php
define( 'MEDIAWIKI', true );

require "mediawiki/includes/GlobalFunctions.php";

echo 'Expected <AAA^BBB"CCC^DDD"EEE^FFF> <GGG>' . "\n";
echo wfShellExec("myecho " . wfEscapeShellArg('AAA^BBB"CCC^DDD"EEE^FFF', 'GGG'));
echo "\n\n\n\n";
echo 'Expected <AAA\\"BB>' . "\n";
echo wfShellExec("myecho " . wfEscapeShellArg('AAA\\"BB'));

I think it is fixed by r69732.