Issue #1677: Handle better a race condition between the interactive interpreter and

the Ctrl-C signal handler on Windows
This commit is contained in:
Tim Golden 2012-06-29 18:27:08 +01:00
parent 5b5619f717
commit b92b757eed
1 changed files with 15 additions and 22 deletions

View File

@ -36,6 +36,7 @@ static int
my_fgets(char *buf, int len, FILE *fp)
{
char *p;
int i;
int err;
while (1) {
if (PyOS_InputHook != NULL)
@ -50,32 +51,24 @@ my_fgets(char *buf, int len, FILE *fp)
return 0; /* No error */
err = errno;
#ifdef MS_WINDOWS
/* In the case of a Ctrl+C or some other external event
interrupting the operation:
Win2k/NT: ERROR_OPERATION_ABORTED is the most recent Win32
error code (and feof() returns TRUE).
Win9x: Ctrl+C seems to have no effect on fgets() returning
early - the signal handler is called, but the fgets()
only returns "normally" (ie, when Enter hit or feof())
/* Ctrl-C anywhere on the line or Ctrl-Z if the only character
on a line will set ERROR_OPERATION_ABORTED. Under normal
circumstances Ctrl-C will also have caused the SIGINT handler
to fire. This signal fires in another thread and is not
guaranteed to have occurred before this point in the code.
Therefore: check in a small loop to see if the trigger has
fired, in which case assume this is a Ctrl-C event. If it
hasn't fired within 10ms assume that this is a Ctrl-Z on its
own or that the signal isn't going to fire for some other
reason and drop through to check for EOF.
*/
if (GetLastError()==ERROR_OPERATION_ABORTED) {
/* Signals come asynchronously, so we sleep a brief
moment before checking if the handler has been
triggered (we cant just return 1 before the
signal handler has been called, as the later
signal may be treated as a separate interrupt).
*/
for (i = 0; i < 10; i++) {
if (PyOS_InterruptOccurred())
return 1;
Sleep(1);
if (PyOS_InterruptOccurred()) {
return 1; /* Interrupt */
}
/* Either the sleep wasn't long enough (need a
short loop retrying?) or not interrupted at all
(in which case we should revisit the whole thing!)
Logging some warning would be nice. assert is not
viable as under the debugger, the various dialogs
mean the condition is not true.
*/
}
#endif /* MS_WINDOWS */
if (feof(fp)) {