quotation issue (Was: broken sed)

[roytam1]

busybox-w32 sed don't work here:

# sed -n "s/^[^#]*ENC.*([^,]*, *\([^,]*\)\(,.*\)*).*/\1_encoder/p" ./libavcodec/allcodecs.c
#

allcodecs.c is here:
http://git.libav.org/?p=libav.git;a=blob_plain;f=libavcodec/allcodecs.c;hb=HEAD

[roytam1]

but sed works when in cmd.exe or msys-mksh calling busybox sed directly, i.e. busybox sed -n "s/^[^#]*ENC.*([^,]*, *\([^,]*\)\(,.*\)*).*/\1_encoder/p" ./libavcodec/allcodecs.c
so the problem should be in busybox sh side.

[roytam1]

I tried to debug it by adding printfs in sed_main:

diff --git a/editors/sed.c b/editors/sed.c
index 070af61..b7817f5 100644
--- a/editors/sed.c
+++ b/editors/sed.c
@@ -1425,6 +1425,10 @@ int sed_main(int argc UNUSED_PARAM, char **argv)

        INIT_G();

+       for(opt=0;opt<argc;opt++)
+               fprintf(stderr, "'%s' ", argv[opt]);
+       fprintf(stderr, "\n");
+
        /* destroy command strings on exit */
        if (ENABLE_FEATURE_CLEAN_UP) atexit(sed_free_and_close_stuff);

from cmd.exe: 'sed' '-n' 's/^[^#]*ENC.*([^,]*, *\([^,]*\)\(,.*\)*).*/\1_encoder/p' './libavcodec/allcodecs.c'
from busybox sh: 'sed' '-n' 's/^[^#]*ENC.*([^,]*, *\\([^,]*\\)\\(,.*\\)*).*/\\1_encoder/p' './libavcodec/allcodecs.c'

so busybox sh adding slashes when it is not needed.

[roytam1]

so the problem is in quote_arg(). It add quotes to arg head and tail, and addslash, why doing both of them in same time?
following patch disable addslashing.

diff --git a/win32/process.c b/win32/process.c
index 53ad44a..cbd5e6d 100644
--- a/win32/process.c
+++ b/win32/process.c
@@ -146,7 +146,7 @@ quote_arg(const char *arg)
        while (*arg) {
                if (*arg == '"')
                        *d++ = '\\';
-               else if (*arg == '\\') {
+               /*else if (*arg == '\\') {
                        int count = 0;
                        while (*arg == '\\') {
                                count++;
@@ -157,7 +157,7 @@ quote_arg(const char *arg)
                        if (*arg == '"') {
                                *d++ = '\\';
                        }
-               }
+               }*/
                *d++ = *arg++;
        }
        *d++ = '"';

[rmyorston]

Windows' handling of command line arguments is really strange. It seems to me that:

• We only need to quote an argument if it contains whitespace.
• There's no point in quoting ? or * because Windows still expands them. Anyhow, if the command line comes from a BusyBox shell it will usually have handled the globbing, so we should turn off Windows globbing for any wildcard that has made it through the shell without expansion. This means that BusyBox commands run from cmd.exe won't do wildcard expansion, though.
• Backslashes only need to be doubled if they occur immediately before double quotes.
• Double quotes need to be escaped with a backslash.

Patch here: http://intgat.tigress.co.uk/rmy/busybox/arg.patch

[roytam1]

Patch here: http://intgat.tigress.co.uk/rmy/busybox/arg.patch

yeah, it works fine here.

[rmyorston]

Patch applied in commit 239bba0