Should all A functions forward to their respective W's?

[James Hawkins]

Hi,

In the current state of wine, we have several A/W functions. 
Sometimes both the A and W functions are separately implemented with
an ansi and unicode implementation respectively.  Other A/W functions
have the A forward to the W, converting the ansi to unicode.  For all
the functions that can, should we forward all A to their W
counterparts?  When it comes to the conformance test suite, this would
be ideal.  Only the A functions would have to be tested and in doing
so, we test the A/W conversion and the functionality of the W
functions (we're striving for all-unicode internally anyway).  This
would reduce the number of bugs, and the time it takes to fix current
bugs.  When both A and W are implemented and we find a bug in one of
them, we have to remember to fix the same bug in the other function. 
For most of the functions, converting ansi to unicode is boilerplate
code.  This process could even be a janitorial project.  What do you
think?

-- 
James Hawkins

[Shachar Shemesh]

James Hawkins wrote:
> Hi,
>
> In the current state of wine, we have several A/W functions. 
> Sometimes both the A and W functions are separately implemented with
> an ansi and unicode implementation respectively.  Other A/W functions
> have the A forward to the W, converting the ansi to unicode.  For all
> the functions that can, should we forward all A to their W
> counterparts?  When it comes to the conformance test suite, this would
> be ideal.  Only the A functions would have to be tested and in doing
> so, we test the A/W conversion and the functionality of the W
> functions (we're striving for all-unicode internally anyway).  This
> would reduce the number of bugs, and the time it takes to fix current
> bugs.  When both A and W are implemented and we find a bug in one of
> them, we have to remember to fix the same bug in the other function. 
> For most of the functions, converting ansi to unicode is boilerplate
> code.  This process could even be a janitorial project.  What do you
> think?

I think that for all functions we can, but care should be taken about 
what "can" means.

For example, some experiments on Windows lead me to believe that 
GetCharacterPlacementW calls the A version, instead of vice versa (that 
is, on Windows 2000). Trying to reorder a string with Unicode characters 
not in the locale simply doesn't work. Obviously, is almost no one calls 
GetCharacterPlacement at all, this is not something we should be overly 
worried about, but I'm just trying to point out that such situations exist.

Other places where converting the dependent calls have been tricky is 
with all the functions that put up a dialog box (and thus require a 
message handler function). These are fairly tricky to get right without 
repeating code, and in many places we preferred not to.

What I'm trying to say is that I'm with you on that one, but stating it 
as you have may lead some people to be overly enthusiastic about things.

          Shachar

-- 
Shachar Shemesh
Lingnu Open Source Consulting ltd.
Have you backed up today's work? http://www.lingnu.com/backup.html

[James Hawkins]

On Apr 5, 2005 1:19 AM, Shachar Shemesh <wine-devel(a)shemesh.biz> wrote:
> I think that for all functions we can, but care should be taken about
> what "can" means.

It's understandable that there will be functions where this process is
not applicable, but for a majority of the api, we can implement the
conversion with relative ease.  Figuring out which conversions would
be easy is up to us I guess.

-- 
James Hawkins

[Uwe Bonnes]

> > > > > "James" == James Hawkins <truiken(a)gmail.com> writes:

James> Hi, In the current state of wine, we have several A/W functions.
    James> Sometimes both the A and W functions are separately implemented
    James> with an ansi and unicode implementation respectively.  Other A/W
    James> functions have the A forward to the W, converting the ansi to
    James> unicode.  For all the functions that can, should we forward all A
    James> to their W counterparts?  When it comes to the conformance test
    James> suite, this would be ideal.  Only the A functions would have to
    James> be tested and in doing so, we test the A/W conversion and the
    James> functionality of the W functions (we're striving for all-unicode
    James> internally anyway).  This would reduce the number of bugs, and
    James> the time it takes to fix current bugs.  When both A and W are
    James> implemented and we find a bug in one of them, we have to remember
    James> to fix the same bug in the other function.  For most of the
    James> functions, converting ansi to unicode is boilerplate code.  This
    James> process could even be a janitorial project.  What do you think?

Even in the case when all A-functions are forwarded to W-functions, a
test-suite for only the A-functions may not be enough, as often there are
some decisions in the A-function before the W-function is called. It the
breakout condition is met, the W-function is not called, and so this
condition is not tested for the W-function


-- 
Uwe Bonnes                bon(a)elektron.ikp.physik.tu-darmstadt.de

Institut fuer Kernphysik  Schlossgartenstrasse 9  64289 Darmstadt
--------- Tel. [phone] -------- Fax. [phone] ----------