553268 - windows _snprintf() may not work like VMPI_snprintf expects it to

Status: VERIFIED FIXED

(Tamarin Graveyard :: Virtual Machine, defect, P2)

Description

[Edwin Smith]

From Nicholas Nethercote nanojit developer:

I just learnt the hard way that snprintf() doesn't exist on Windows.
There is _snprintf(), but it's unsafe -- it doesn't guarantee
terminating with a NUL char.  sprintf_s() is the function that does
that.

http://msdn.microsoft.com/en-us/library/2ts7cx93%28VS.80%29.aspx
http://msdn.microsoft.com/en-us/library/ce3zzk1k%28VS.80%29.aspx

I mention it because I see the following lines in TR:

./platform/win32/win32-platform.h:#define VMPI_snprintf		::_snprintf
./VMPI/SpyUtilsWin.cpp:	#pragma warning(disable: 4995) //disabling
warning for deprecated _snprintf
./VMPI/SpyUtilsWinMo.cpp:	#pragma warning(disable: 4995) //disabling
warning for deprecated _snprintf

Comment 1

[Lars T Hansen]

Right.  The problem is that _snprintf_s does more than just NUL-terminate, it actually invokes an exception handler if the buffer can't accomodate the output.  We probably don't want that.

Comment 2

[Lars T Hansen]

Attachment: Patch

Implements VMPI_snprintf and VMPI_vsnprintf as functions on Windows, with reasonable semantics.

Removes the warning-disabling pragmas for _snprintf usage.

Adds VMPI_vsnprintf to all the platform header files (only Windows had it).

Changes some uses in the eval code of non-VMPI functions to use VMPI functions - fallout from removing a use of _snprintf.

Comment 3

[Lars T Hansen]

A selftest test case would be handy here.

Comment 4

[Lars T Hansen]

tamarin-redux-argo changeset:   3846:1bc40e7217a8
tamarin-redux changeset:   4109:0e6e60921250

Comment 5

[Brent Baker]

Looks like this patch does not compile properly on windows mobile:

MMgc/GCLog.cpp(49) : error C3861: 'vsnprintf': identifier not found

Comment 6

[Lars T Hansen]

(In reply to comment #5)
> Looks like this patch does not compile properly on windows mobile:
> 
> MMgc/GCLog.cpp(49) : error C3861: 'vsnprintf': identifier not found

Bummer for Windows Mobile.

Comment 7

[Lars T Hansen]

Attachment: Selftests for VMPI_snprintf

These test basic functionality of VMPI_snprintf:

- formatting various types
- handling buffer overflow and termination
- return values

To the best of my knowledge I'm going after only ISO C functionality here.

Comment 8

[Lars T Hansen]

Attachment: Turn two calls to vsnprintf into calls to VMPI_vsnprintf

Comment 9

[Lars T Hansen]

Attachment: Reasonably conformant implementation of VMPI_vsnprintf

Performs a rough-and-ready parse of the format string and hands individual problems off to _snprintf, and assembles a string from the results.  Unlike _snprintf and _vsnprintf it returns the number of chars that would have been printed had the buffer been large enough.

Does not quite obsolete patch #433597, but replaces part of it.

Comment 10

[Lars T Hansen]

Comment on attachment 434377 [details] [diff] [review]
Turn two calls to vsnprintf into calls to VMPI_vsnprintf

Moved to bug #554659.

Comment 11

[Lars T Hansen]

Comment on attachment 434375 [details] [diff] [review]
Selftests for VMPI_snprintf

Cancelling review as patch will be moved.

Comment 12

[Lars T Hansen]

Comment on attachment 434380 [details] [diff] [review]
Reasonably conformant implementation of VMPI_vsnprintf

Cancelling review as patch will be moved.

Comment 13

[Lars T Hansen]

Comment on attachment 434375 [details] [diff] [review]
Selftests for VMPI_snprintf

Moved to bug #554664

Comment 14

[Lars T Hansen]

Comment on attachment 434380 [details] [diff] [review]
Reasonably conformant implementation of VMPI_vsnprintf

Moved to bug #554664

Comment 15

[Lars T Hansen]

(In reply to comment #5)
> Looks like this patch does not compile properly on windows mobile:
> 
> MMgc/GCLog.cpp(49) : error C3861: 'vsnprintf': identifier not found

That issue is now tracked in bug #554659 as it turned out to be more complicated.

Comment 16

[Lars T Hansen]

Test cases have been moved to bug #554664.