- Wine-Devel - list.winehq.org

Standard controls
by Dimitrie O. Paun Sept. 11, 2002

Sept. 11, 2002

And since I feel frisky lately, I want to run another idea by you guys. We have the standard controls bundled with USER, just like Windows. In fact, it seems XP bundles them with the common control, but that's another story. My point is, that there's no logical reason they should be in USER. In fact, in the future we may want to provide other implementations which integrate better with GNOME and/or KDE. So what about moving them to a separate DLL, say winectrl? -- Dimi.

2 2

winebuild & (name, heap, init)
by Dimitrie O. Paun Sept. 11, 2002

Sept. 11, 2002

Alexandre, Why don't we take the following directives: name, heap, init out of the .spec file, and put them in the Makefile (transfered to winebuild through command line args)? Some of them are supported in .DEF file as well, but all of them are supported as arguments to LINK. I think it will gain us flexibility if we also support them as command line arguments. With that in mind, it would be neat if we could specify a flag at configure time to build most of Wine as PE with a cross mingw. It would be way cool if I could take the same comctl32.dll I have on Linux, copy it on my Windows machine, and test it out... Anyway, what about a patch to at least support all of this as command line options as well? -- Dimi.

4 7

RE: winebuild & (name, heap, init)
by Patrik Stridvall Sept. 11, 2002

Sept. 11, 2002

> On September 11, 2002 11:33 am, Patrik Stridvall wrote: > > Sure, but I have written the Perl code I know how ad hoc it is. > > It works under reasonable assumptions however beyond that... > > But we don't *want* Wine's sources to look beyond reasonable. > I see no problem. Even better, if this helps us detect unreasonable > looking sources... Well winapi_check already does that so no additional advantage...

1 0

RE: winebuild & (name, heap, init)
by Patrik Stridvall Sept. 11, 2002

Sept. 11, 2002

> -----Original Message----- > From: Dimitrie O. Paun [mailto:dpaun(a)rogers.com] > Sent: Wednesday, September 11, 2002 5:37 PM > To: Patrik Stridvall; Francois Gouget > Cc: Wine Devel > Subject: Re: winebuild & (name, heap, init) > > > On September 11, 2002 11:24 am, Patrik Stridvall wrote: > > > BTW, how many are IO? I would suspect most are O only... > > > > For strings, not that many. Can't think of any right now. > > However there are quite a few LPDWORD that specifies the > > allocated length of the output buffer and the function > > is should write the used length (or needed length, > > if the function faild) in the pointed to DWORD. > > Dude, you're killing me! I often do that yet you still live. :-) > We care to distinguish I/O for strings _only_. And since we > don't have any, we have no problem!!! We have I found an example: LPTSTR PathAddBackslash(LPTSTR lpszPath); A few more of the Path functions are like that IIRC. Then there might be more function. The Windows API is larger we have only begun to implement it... > For LPDWORD, print them > always, they are fixed length, you (as the guy looking at the > traces) know if you care about them or not... True. How if you wish to automatically compare debug output diffs you have a problem. Sure pointer and such usually differs but that can be fixed with some sort of mapping table. However the value of unintialized variables are "random" or rather non-derministic since they might depend of subtle differences in the enviroment. But I do agree that strings is the larger problem...

1 0

RE: winebuild & (name, heap, init)
by Patrik Stridvall Sept. 11, 2002

Sept. 11, 2002

> Were we distinguishing IN/OUT and OUT before in the spec file? No. I suggest to Alexandre that we should add such information but he rejected my patch...

1 0

RE: winebuild & (name, heap, init)
by Jeff Smith Sept. 11, 2002

Sept. 11, 2002

1 0

RE: winebuild & (name, heap, init)
by Patrik Stridvall Sept. 11, 2002

Sept. 11, 2002

> On September 11, 2002 10:16 am, Patrik Stridvall wrote: > > #define HELPER(x,y,z) DWORD WINAPI x(y arg1, z *arg1) > > > > FOO(Bar, INT, HWND) { > > /* ... */ > > } > > > > But we don't need to support that. If it doesn't get recognized, > the function is not exported. You can manually look in the generated > spec file and see it's not there. > > I don't think we need to be too clever. Just get the 80% that can be > done with 20% of effort. Well as I said "winapi_extract --spec-files" works to some extent even now and getting it to work good enough might not be that much effort. However I'm worried about the "stabillity" of the result. > It's a temporaray solution anyway: we need > help from the compiler, writing a C parser (no, a gcc-C parser) is > a lot of work. Yes. I know I have tried. :-) > Mingw supports __declspec, Urlich, what would it take > for the regular gcc to support it under Linux/ELF? But what should it do? Just generate a special dllexport section that you can read with appropriate tools or what? > But beyond that, we can *today* compile Wine with Mingw as PE. So > what we need is a checkspec make target which would run your script, > compile Wine as PE, extract what needs extracting from there, and > compare results. This does not need to happen all the time, only > from time to time by some people that (1) have Mingw installed, > and (2) care to do it :). True. However I still a little worried about the "stabillity" of the solution.

2 1

RE: winebuild & (name, heap, init)
by Patrik Stridvall Sept. 11, 2002

Sept. 11, 2002

> On September 11, 2002 05:48 am, Francois Gouget wrote: > > Hmm, for strings the spec files currently distinguish between input > > strings (str) and output strings (ptr). This cannot be > extracted from > > the source files... unless we start using IN/OUT macros everywhere? > > It's all in the types, dude... > > LPCSTR/LPCWSTR = IN, LPSTR/LPWSTR=IN/OUT. The is almost always correct. Unfortunately Microsoft is not consequent in the regard, but then I guess that we could modify the API slightly. If a WineLib application breaks, it was in some meaning broken anyway. However how do know whether it is IN/OUT or OUT? For debug messages that is quite important since the pointer in the OUT case might point to uninitialized memory which shouldn't be printed in the debug output...

2 1

RE: winebuild & (name, heap, init)
by Patrik Stridvall Sept. 11, 2002

Sept. 11, 2002

> On September 11, 2002 06:18 am, Patrik Stridvall wrote: > > However I might be able to write a very specialized parser, that > > pipes the .c files through the C preprosessor, that work all of the > > time, but this is quite some work as well. > > But we shouldn't need that. Wine has a very good strucutre right now, > and were that's broken, we can fix... Sure. But what would tell you that it was broken? winapi_check currently, among other thing, compares the .spec files with the source code. When it reports something there are three possibillities 1. winapi_check is broken (because the code doesn't have good structure) 2. the .spec file is broken 3. the .c file is broken Currently I check manually and I try to make an educated guess of which is the cases and then fix it. If you extract the information from the source you don't have that luxury. If it misses a function or have some other bug you have no backup. Since the .spec files exists winapi_check knows what functions it is expected to find in the source code and if is doesn't it reports it. So if for example we have function Foo that looks like. #define HELPER(x,y,z) DWORD WINAPI x(y arg1, z *arg1) FOO(Bar, INT, HWND) { /* ... */ } winapi_check will miss it (since it in an ad hoc parser), but since Bar is declared in some .spec file it will complain that Bar was never found. Then I manually search the source, and try to fix the problem somehow. If you on the other hand extract the .spec files (of .def files) from the source the function Bar would be missed and you would never detect it... That is why the parser needs to be so good that it "always" works.

2 1

RE: winebuild & (name, heap, init)
by Patrik Stridvall Sept. 11, 2002

Sept. 11, 2002

> On September 10, 2002 02:15 pm, Alexandre Julliard wrote: > > > Well, the problem is that the __declspec stuff is usually inside > > macros, so you have to run the code through the C > preprocessor to find > > them. It can be done, but it's not that trivial. Also one thing you > > need to extract is the relay debugging information (and for 16-bit > > code information for the relay routines), and this requires > a C parser > > (or some of Patrik's magic Perl stuff). > > I'm sure if we have the political will at the top, we'll find a > solution... :) Patrik, care to help us extract the required stuff > directly out of the source file? Sure. BTW winapi_extract can already do that: winapi_extract --spec-files This generates a .spec2 file that should hopefully be the same as the .spec file. It currently isn't (except for a few files) though but that can hopefully be fixed. Outputing in another format like .def files or whatever should be quite easy. However note that winapi_extract is quite ad hoc and specfic for parsing the Wine source tree. It is not a generalized C parser. I have begun with a generalized C parser in winapi_fixup however it doesn't quite work that good yet. Parsing generalized C is non trivial. > > A related idea I had is to store the relay debugging independently > > from the .spec file; you could then use it in other cases, notably > > with snoop debugging of native dlls. And then you no longer need a > > .spec, you can use a .def or directly extract the __declspecs, at > > least for Win32. This would also potentially allow more intelligent > > relay routines, maybe even per-dll relay code that knows > about the dll > > internal data structures; and hopefully also relay debugging of COM > > interfaces. > > Exactly! That's where I was driving at. But this depends on getting > stuff automatically out of .c file, because this way we can retain > the type information, which we now lose (except for strings). Once > we have the type information, the sky's the limit... But most > imprtantly, > we eliminate (1) a potential source of mistakes, (2) a barrier for > new commers (which are likely to be from the Windows world), and > (3) avoid maintenaning the 700KB of .specs!!! Agreed, but the script that does it is also a potential source of mistake and the current state of the winapi tools are that they work MOST of the time. However if we are to replace the .spec files it must work ALL the time. However I might be able to write a very specialized parser, that pipes the .c files through the C preprosessor, that work all of the time, but this is quite some work as well. Note however that piping every source file through the C prepropessor is VERY SLOW. That is one of the reason the winapi tools doesn't do that.

2 1