[PATCH] dlls/kernelbase, programs/conhost: don't delay the TTY input thread for PE executables

[Eric Pouech]

currently, conhost delays the start of the input thread until the first
console read operation

some PE executables (like mingw's gdb port) just do something like:
- WaitForSingleObject(GetStdHandle(STD_INPUT_HANDLE), INFINITE) and hang
  for ever (the read operations are done *after* the wait operation
  succeeds)
(of course, the real wait operation is more complex, but the problematic
part boils down to that)

I understand that the delay has been introduced so that executables still
reading directly from the unix fd will continue to work.

But I could miss some other use cases.

This patch forces the start the tty input thread in conhost if current
executable is a PE executable.

This lets mingw's gdb work a bit further (at least allowing some user input
<g>) under wine. (x86_64 seems to be quite ok, i686 still needs come
cudling)

Signed-off-by: Eric Pouech

---
 dlls/kernelbase/console.c  |    8 ++++++--
 programs/conhost/conhost.c |    9 +++++++--
 2 files changed, 13 insertions(+), 4 deletions(-)

diff --git a/dlls/kernelbase/console.c b/dlls/kernelbase/console.c
index a7eeb439232..b6f2c4fb394 100644
--- a/dlls/kernelbase/console.c
+++ b/dlls/kernelbase/console.c
@@ -1694,8 +1694,12 @@ static HANDLE create_pseudo_console( COORD size, HANDLE input, HANDLE output, HA
     }
     else
     {
-        swprintf( cmd, ARRAY_SIZE(cmd), L""%s" --unix --width %u --height %u --server 0x%x",
-                  conhost_path, size.X, size.Y, server );
+        LDR_DATA_TABLE_ENTRY *ldr;
+        const WCHAR* start_thread = L"";
+        if (LdrFindEntryForAddress( GetModuleHandleW( NULL) , &ldr ) == STATUS_SUCCESS &&
+            !(ldr->Flags & LDR_WINE_INTERNAL)) start_thread = L"--thread ";
+        swprintf( cmd, ARRAY_SIZE(cmd), L""%s" --unix %s--width %u --height %u --server 0x%x",
+                  conhost_path, start_thread, size.X, size.Y, server );
     }
     Wow64DisableWow64FsRedirection( &redir );
     res = CreateProcessW( conhost_path, cmd, NULL, NULL, TRUE, DETACHED_PROCESS, NULL, NULL,
diff --git a/programs/conhost/conhost.c b/programs/conhost/conhost.c
index 05447d52dd5..27788fbd6a8 100644
--- a/programs/conhost/conhost.c
+++ b/programs/conhost/conhost.c
@@ -2736,7 +2736,7 @@ static int main_loop( struct console *console, HANDLE signal )

int __cdecl wmain(int argc, WCHAR *argv[])
 {
-    int headless = 0, i, width = 0, height = 0;
+    int headless = 0, i, width = 0, height = 0, start_thread = 0;
     HANDLE signal = NULL;
     WCHAR *end;

@@ -2765,6 +2765,11 @@ int __cdecl wmain(int argc, WCHAR *argv[])
             headless = 1;
             continue;
         }
+        if (!wcscmp( argv[i], L"--thread"))
+        {
+            start_thread = 1;
+            continue;
+        }
         if (!wcscmp( argv[i], L"--width" ))
         {
             if (++i == argc) return 1;
@@ -2812,7 +2817,7 @@ int __cdecl wmain(int argc, WCHAR *argv[])
         console.tty_input  = GetStdHandle( STD_INPUT_HANDLE );
         console.tty_output = GetStdHandle( STD_OUTPUT_HANDLE );
         init_tty_output( &console );
-        if (!console.is_unix && !ensure_tty_input_thread( &console )) return 1;
+        if ((!console.is_unix || start_thread) && !ensure_tty_input_thread( &console )) return 1;
     }
     else
     {

[Jacek Caban]

Hi Eric,

On 12/6/21 3:59 PM, Eric Pouech wrote:
> currently, conhost delays the start of the input thread until the first
> console read operation
>
> some PE executables (like mingw's gdb port) just do something like:
> - WaitForSingleObject(GetStdHandle(STD_INPUT_HANDLE), INFINITE) and hang
>  for ever (the read operations are done*after*  the wait operation
>  succeeds)
>
> (of course, the real wait operation is more complex, but the problematic
> part boils down to that)
>
> I understand that the delay has been introduced so that executables still
> reading directly from the unix fd will continue to work.
>
> But I could miss some other use cases.

The idea is that it's better to not change console parameters 
(IOCTL_CONDRV_SETUP_INPUT) unless we need to. Majority of applications 
never read from console and this avoids changing current console mode 
every time you run Wine.
> This patch forces the start the tty input thread in conhost if current
> executable is a PE executable.

Maybe we could detect the problem in server somewhere around 
console_input_add_queue and send an ioctl to conhost process to 
initialize input thread.
> This lets mingw's gdb work a bit further (at least allowing some user input
> <g>) under wine. (x86_64 seems to be quite ok, i686 still needs come
> cudling)

Nice :)

Thanks,

Jacek