[PATCH 0/1] MR7586: cmd: Handle cntl-c events correctly

[Jeff Latimer (＠greyghoster)]

https://bugs.winehq.org/show_bug.cgi?id=27605 describes how cntl-c is not handled well in cmd.exe, this patch allows the cntl-c to be returned in the buffer so that it can processed.

[PATCH 1/1] cmd: Handle cntl-c events correctly

[Jeff Latimer]

From: Jeff Latimer

---
 programs/cmd/batch.c | 18 +++++++++++++++++-
 1 file changed, 17 insertions(+), 1 deletion(-)

diff --git a/programs/cmd/batch.c b/programs/cmd/batch.c
index 20ae894dbfd..1cd883c0621 100644
--- a/programs/cmd/batch.c
+++ b/programs/cmd/batch.c
@@ -223,17 +223,32 @@ WCHAR *WCMD_fgets(WCHAR *buf, DWORD noChars, HANDLE h)
   DWORD charsRead;
   BOOL status;
   DWORD i;
+  DWORD conmode, oldconmode;
+  CONSOLE_READCONSOLE_CONTROL control;
+
+  GetConsoleMode(h, &oldconmode);
+  conmode = oldconmode & ~ENABLE_PROCESSED_INPUT;
+  SetConsoleMode(h, conmode);
+  control.nLength = sizeof(CONSOLE_READCONSOLE_CONTROL);
+  control.nInitialChars = 0;
+  control.dwCtrlWakeupMask = 0x0008;
+  control.dwConsoleKeyState = 0;

/* We can't use the native f* functions because of the filename syntax differences
      between DOS and Unix. Also need to lose the LF (or CRLF) from the line. */

-  if (VerifyConsoleIoHandle(h) && ReadConsoleW(h, buf, noChars, &charsRead, NULL) && charsRead) {
+  if (VerifyConsoleIoHandle(h) && ReadConsoleW(h, buf, noChars, &charsRead, &control) && charsRead) {
+      SetConsoleMode(h, oldconmode);
       if (!charsRead) return NULL;

/* Find first EOL */
       for (i = 0; i < charsRead; i++) {
           if (buf[i] == '\n' || buf[i] == '\r')
               break;
+          if (buf[i] == 0x03) {
+             WCMD_output(L"^C\n");
+             i = 0;
+             break;}
       }
   }
   else {
@@ -242,6 +257,7 @@ WCHAR *WCMD_fgets(WCHAR *buf, DWORD noChars, HANDLE h)
       UINT cp;
       const char *p;

+      SetConsoleMode(h, oldconmode);
       cp = GetOEMCP();
       bufA = xalloc(noChars);

[eric pouech (＠epo)]

eric pouech (@epo) commented about programs/cmd/batch.c:
> DWORD charsRead;
>    BOOL status;
>    DWORD i;
> - DWORD conmode, oldconmode;
> - CONSOLE_READCONSOLE_CONTROL control;
> -
> - GetConsoleMode(h, &oldconmode);

I think this would be better handled by rewriting the code as:

`if (GetConsoleMode(h, &oldmode)) {`

`/* put all console related code here */`

`SetConsoleMode(h, oldmode);`

`}`

`else`

`{`

`/* other handles (file, pipe) */`

[eric pouech (＠epo)]

eric pouech (@epo) commented about programs/cmd/batch.c:
> DWORD i;
> -
>
> DWORD conmode, oldconmode;
> -
>
> CONSOLE_READCONSOLE_CONTROL control;
> -
> -
>
> GetConsoleMode(h, &oldconmode);
> -
>
> conmode = oldconmode & ~ENABLE_PROCESSED_INPUT;
> -
>
> SetConsoleMode(h, conmode);
> -
>
> control.nLength = sizeof(CONSOLE_READCONSOLE_CONTROL);
> -
>
> control.nInitialChars = 0;
> -
>
> control.dwCtrlWakeupMask = 0x0008;
> -
>
> control.dwConsoleKeyState = 0;
>
> /* We can't use the native f* functions because of the filename syntax differences
>    between DOS and Unix. Also need to lose the LF (or CRLF) from the line. */
> - if (VerifyConsoleIoHandle(h) && ReadConsoleW(h, buf, noChars, &charsRead, NULL) && charsRead) {

we can get rid of VerifyConsoleHandle() as GetConsoleMode() succeeded

[eric pouech (＠epo)]

eric pouech (@epo) commented about programs/cmd/batch.c:
> -
>
> control.dwConsoleKeyState = 0;
>
> /* We can't use the native f* functions because of the filename syntax differences
>    between DOS and Unix. Also need to lose the LF (or CRLF) from the line. */
> - if (VerifyConsoleIoHandle(h) && ReadConsoleW(h, buf, noChars, &charsRead, NULL) && charsRead) {
> -
>
> if (VerifyConsoleIoHandle(h) && ReadConsoleW(h, buf, noChars, &charsRead, &control) && charsRead) {
> -
>  SetConsoleMode(h, oldconmode);
>  if (!charsRead) return NULL;
>
>  /* Find first EOL */
>  for (i = 0; i < charsRead; i++) {
>      if (buf[i] == '\n' || buf[i] == '\r')
>          break;
> -
>      if (buf[i] == 0x03) {
> -
>         WCMD_output(L"^C\n");

I don't see ^C printed on native (Win10)

[eric pouech (＠epo)]

eric pouech (@epo) commented about programs/cmd/batch.c:
> DWORD charsRead;
>    BOOL status;
>    DWORD i;
> - DWORD conmode, oldconmode;
> - CONSOLE_READCONSOLE_CONTROL control;
> -
> - GetConsoleMode(h, &oldconmode);
> - conmode = oldconmode & ~ENABLE_PROCESSED_INPUT;
> - SetConsoleMode(h, conmode);
> - control.nLength = sizeof(CONSOLE_READCONSOLE_CONTROL);
> - control.nInitialChars = 0;
> - control.dwCtrlWakeupMask = 0x0008;

perhaps use 1u << 0x03 instead (to make it clearer that the intent in ctrl-c) (note one day we might even add tab/shift tab here)

[eric pouech (＠epo)]

eric pouech (@epo) commented about programs/cmd/batch.c:
> /* We can't use the native f* functions because of the filename syntax differences
>       between DOS and Unix. Also need to lose the LF (or CRLF) from the line. */
> - if (VerifyConsoleIoHandle(h) && ReadConsoleW(h, buf, noChars, &charsRead, NULL) && charsRead) {
> -
>
> if (VerifyConsoleIoHandle(h) && ReadConsoleW(h, buf, noChars, &charsRead, &control) && charsRead) {
> -
>  SetConsoleMode(h, oldconmode);
>  if (!charsRead) return NULL;
>
>  /* Find first EOL */
>  for (i = 0; i < charsRead; i++) {
>      if (buf[i] == '\n' || buf[i] == '\r')
>          break;
> -
>      if (buf[i] == 0x03) {
> -
>         WCMD_output(L"^C\n");
> -
>         i = 0;
> -
>         break;}

misplaced curly bracket

[eric pouech (＠epo)]

only works as expected under user32 console

but complely breaks the unix console

culprit is not the patch, is that conhost is closed with ctrl-c when processed input is turned off

(without the patch on unix console, there are unwanted cursor movement, but after hitting ctrl-c one can still do something; with the patch cmd is "frozen")

there at least two issues to consider:

1. ReadConsole with control on unix console doesn't report ctrl-c
2. cmd shouldn't freeze when the console has been closed

either one of 1 or 2 must be supplied before accepting this patch

apart from that, mostly stylistic remarks

[Jeff Latimer (＠greyghoster)]

On Thu Mar 20 00:49:02 2025 +0000, eric pouech wrote:
> only works as expected under user32 console
> but complely breaks the unix console
> culprit is not the patch, is that conhost is closed with ctrl-c when
> processed input is turned off
> (without the patch on unix console, there are unwanted cursor movement,
> but after hitting ctrl-c one can still do something; with the patch cmd
> is "frozen")
> there at least two issues to consider:
> - ReadConsole with control on unix console doesn't report ctrl-c
> - cmd shouldn't freeze when the console has been closed
>
> either one of 1 or 2 must be supplied before accepting this patch
> apart from that, mostly stylistic remarks

Thanks, I will look more closely at it.