[Bug 53172] New: advapi32:registry - test_enum_value() has a pair of rare failures in UTF-8 system locales

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=53172

Bug ID: 53172
           Summary: advapi32:registry - test_enum_value() has a pair of
                    rare failures in UTF-8 system locales
           Product: Wine
           Version: unspecified
          Hardware: x86-64
                OS: Windows
            Status: NEW
          Severity: normal
          Priority: P2
         Component: advapi32
          Assignee: [e-mail]
          Reporter: [e-mail]

advapi32:registry - test_enum_value() has a pair of rare failures in UTF-8
system locales:

registry.c:558: Test failed: data set to 'xxxxxxxxxxxxxxxxxxxx' instead of
'foobar' or x's, data_count=21
registry.c:576: Test failed: data set to 'xxxxxxxxxxxxxxxxxxxx' instead of
'foobar' or x's, data_count=21

https://test.winehq.org/data/patterns.html#advapi32:registry

The line 558 and 576 failures happen about with the same frequency (~13% each)
but independently from each other so that most times there is at most one in
the report.

And they only happen in the TestBot's two UTF-8 test configurations:
w10pro64-en-AE-u8 and w10pro64-hi-u8.

[Bug 53172] advapi32:registry - test_enum_value() has a pair of rare failures in UTF-8 system locales

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=53172

François Gouget  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |source, testcase

[Bug 53172] advapi32:registry - test_enum_value() has a pair of rare failures in UTF-8 system locales

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=53172

François Gouget  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|[e-mail]        |[e-mail]

--- Comment #1 from François Gouget  ---
Windows's behavior RegEnumValueA() is broken in all cases where the name buffer
is too small, but particularly so in double-byte and UTF-8 locales.

The test is configured to enumerate a single value:

Test: REG_SZ "foobar"

With:
  data_count = 20
  memset(data, 'x', sizeof(data)) 
  RegEnumValueA(...too small name buffer..., data, &data_count)

We randomly get one of the results below:
* data_count=7
  This is strlen("foobar")+1 which makes sense
  data[data_count] = 0
  Modifying the buffer makes no sense in the first place. But should it be
modified, then it should do data[data_count-1] = 0. Maybe this is a REG_SZ vs.
REG_MULTI_SZ issue.

* 7 < data_count < 20 and data[data_count] = 0
  The required buffer size is 7, there is no reason to claim more is needed.
  The returned values don't even seem to be random; it's mostly 8-11 and 13.
  There is still the data[data_count] vs. data[data_count-1] issue.

* data_count >= 20 and data is unmodified
  This seems to be specific to the UTF-8 cases (w10pro64_en_AE_u8 and
w10pro64_hi_u8).
  Specifically the only value seen in this case is data_count = 21. Even if the
initial data_count value is changed to 16. So why 21???
  And fortunately RegEnumValueA() did not try to set data[data_count] = 0
(would be a buffer overflow).
  This is what's causing the failures for two reasons:
  - We memset(data, 'x', sizeof(data)) which means it's not \0-terminated. So
strspn( data, "x" ) may return anything depending on what comes after the
buffer. At least there's only a 1/256 chance of it being wrong.
  - data[data_count] is also outside the buffer and this time it has a 255/256
chance of being wrong (though in practice it's not random and may be more
likely to be 0).

[Bug 53172] advapi32:registry - test_enum_value() has a pair of rare failures in UTF-8 system locales

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=53172

--- Comment #2 from François Gouget  ---
Sent a merge request:
https://gitlab.winehq.org/wine/wine/-/merge_requests/2324

[Bug 53172] advapi32:registry - test_enum_value() has a pair of rare failures in UTF-8 system locales

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=53172

François Gouget  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
      Fixed by SHA1|                            |95f2cc0f15b1dde0318e75e2a7f
                   |                            |21957fa76af55
         Resolution|---                         |FIXED

--- Comment #3 from François Gouget  ---
This should be fixed:

commit 95f2cc0f15b1dde0318e75e2a7f21957fa76af55
Author: François Gouget 
Date:   Fri Mar 3 16:23:07 2023 +0100

advapi32/tests: Fix the RegEnumValueA() tests in UTF-8 locales.

On Windows RegEnumValueA() is buggy such that when the value name
    buffer is too small, it may way overestimate the required data buffer
    size, sometimes returning more than the sufficiently large buffer it was
    provided, in which case that buffer is untouched and thus not
    NUL-terminated.
    So modify the tests to accept this broken result and avoid checking
    out-of-bounds data.

Wine-Bug: https://bugs.winehq.org//show_bug.cgi?id=53172

[Bug 53172] advapi32:registry - test_enum_value() has a pair of rare failures in UTF-8 system locales

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=53172

Alexandre Julliard  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |CLOSED

--- Comment #4 from Alexandre Julliard  ---
Closing bugs fixed in 8.4.