Re: [ros-dev] Status Meeting (November 2020)

[Hermès BÉLUSCA-MAÏTO]

What about that discussion about why we have ditched out MSVC 2010 support
(starting to depend on that fact, and thus, breaking any possibility of
self-hosting ROS building on ROS using MSVC compiler (without the IDE),
since later versions of MSVC only work on Vista+, and as an indirect
consequence, too, opening the can of worms, that is, being able to
"pollute", involuntarily or not, the ReactOS core code (kernel + drivers)
with C99+ non-standard portable code -- ),
and without having first warned heavily through all the mailing lists etc. ,
thus allowing the developers to voice their concerns publicly in the MLs ?
This was done around in mattermost only, after we decided for some reason
that we wouldn't need the VC2010 buildbots anymore.

Hermes.

-----Original Message-----
From: Ros-dev <ros-dev-bounces(a)reactos.org> On Behalf Of Colin Finck
Sent: Sun Nov 15 17:24:22 UTC 2020
To: ReactOS Development List <ros-dev(a)reactos.org>
Subject: [ros-dev] Status Meeting (November 2020)

Hi all,

With several people asking for a status meeting, let's have another one 
before the year is over.
Let me invite you to the meeting on

       Thursday, 26th November 2020
       19:00 UTC
       Mattermost private channel "Meeting"

So far, there is only one point on the agenda:

* Achievements and Future Outlook (everyone)
What have you been working on and what are your plans?

Please submit further agenda proposals by answering to this mail.


Looking forward to see you!

Colin

[Erkin Alp Güney]

What do you mean by C99+ non-portable code? GCC and Clang have -std=c(++)11
to keep it in an ISO compliant manner.

26 Kas 2020 Per 00:27 tarihinde Hermès BÉLUSCA-MAÏTO <hermes.belusca(a)sfr.fr>
şunu yazdı:

>   What about that discussion about why we have ditched
> out MSVC 2010 support
>  (starting to depend on that fact, and thus, breaking any possibility of
>  self-hosting ROS building on ROS using MSVC compiler (without the IDE),
>  since later versions of MSVC only work on Vista+, and as an indirect
>  consequence, too, opening the can of worms, that is, being able to
>  "pollute", involuntarily or not, the ReactOS core code (kernel + drivers)
>  with C99+ non-standard portable code -- ),
>  and without having first warned heavily through all the mailing lists etc.
>  ,
>  thus allowing the developers to voice their concerns publicly in the MLs ?
>  This was done around in mattermost only, after we decided for some reason
>  that we wouldn't need the VC2010 buildbots anymore.
>
>  Hermes.
>
>  -----Original Message-----
>  From: Ros-dev <ros-dev-bounces(a)reactos.org> On Behalf Of Colin Finck
>  Sent: Sun Nov 15 17:24:22 UTC 2020
>  To: ReactOS Development List <ros-dev(a)reactos.org>
>  Subject: [ros-dev] Status Meeting (November 2020)
>
>  Hi all,
>
>  With several people asking for a status meeting, let's have another one
>  before the year is over.
>  Let me invite you to the meeting on
>
>         Thursday, 26th November 2020
>         19:00 UTC
>         Mattermost private channel "Meeting"
>
>  So far, there is only one point on the agenda:
>
>  * Achievements and Future Outlook (everyone)
>  What have you been working on and what are your plans?
>
>  Please submit further agenda proposals by answering to this mail.
>
>
>  Looking forward to see you!
>
>  Colin
>
>
>  _______________________________________________
>  Ros-dev mailing list
>  Ros-dev(a)reactos.org
>  http://reactos.org/mailman/listinfo/ros-dev
>

[Colin Finck]

Am 25.11.2020 um 22:26 schrieb Hermès BÉLUSCA-MAÏTO:
>   What about that discussion about why we have ditched
> out MSVC 2010 support 
You mean reintroducing an ancient compiler that forces us to hack around 
its incapacity to deal with anything more modern than C89? One that 
already makes it harder to import third-party components and attract new 
developers?

Let's face it, we already make our incredible job of writing a 
Windows-compatible operating system even harder than Microsoft's - by 
supporting both GCC and MSVC. We spend enormous amounts of time to just 
get the basics right for both toolchains, and still fail at things like 
the C++ standard library. Ditching MSVC 2010 finally paves the way for 
importing a modern and trusted third-party C++ library like Clang's 
libc++ to replace our hacked and unmaintained stlport. Going even 
further, we could finally draw from the huge pool of talented C++ 
developers, who leverage the benefits of C++11 and beyond. Think about 
that next time you debug a memory leak or a deadlock that would have 
been impossible through std::unique_ptr or std::mutex.

This project won't ever become usable in a lifetime if we don't go with 
the time but stick to old habits forever. There is no inherent value in 
supporting old compilers, just wasted developer time.
And don't tell me about self-hosting: It's not like anybody productively 
uses ReactOS for building ReactOS right now. If you just want to 
demonstrate that, use the RosBE GCC-based toolchain. I already spent 
hours on backporting that back to ReactOS/Windows XP...


Now I usually would have put that on the agenda for the meeting and 
asked for a vote.
But as this year's ditching of MSVC 2010 is just the implementation of a 
meeting decision from 2 years ago 
(https://reactos.org/project-news/december-2018-meeting-minutes/), 
meeting votes are apparently not considered binding here. Neither is 
apparently the discussion we had in the related PR 
https://github.com/reactos/reactos/pull/2658.


So let's have an open discussion and voting process right here on the 
mailing list for everyone to see, to decide this once and for all!
Just reply to this mail, tell about your reasons - and more importantly 
tell whether you support or reject dropping MSVC 2010 support.

Votes by CORE REACTOS MEMBERS are then counted on 1ST JANUARY 2021.


At this point, it should be obvious that I myself SUPPORT the decision 
to ditch building ReactOS with MSVC 2010.


Cheers,

Colin

Moin,

a few questions:
   * How much work needs to be done to get newer versions of MSVC running 
by e.g. using the shims? (not GUI, only the toolchain, and possibly hack 
around installer problems)?
   * How big is the danger to lose the capability to exchange ReactOS and 
Win 2k3 system files?
   * There isn't a time plan to get at least the core system file 
structure, exports and structures at Win 6+ level yet, right?
   * How big is the danger that newer MSVC versions can't be used to 
compile ReactOS anymore? (E.g. loosing the capability to build Win XP 
compatible files)

I think the first question/solution could be a short time compromise, 
the rest points at the "real" long time solution: targeting a newer 
Windows version..

Best regards,
Michael Fritscher

[Victor Perevertkin]

Hey,

>   How much work needs to be done to get newer versions
> of MSVC
>  running  
This needs our compatibility layer to be finished (it's progressing,
but slowly)

>   How big is the danger to lose the capability to
> exchange ReactOS
>  and Win 2k3 system files? 
As far as I know, we don't have issues with that atm
with the latest
MSVC

>   There isn't a time plan to get at least the core
> system file 
>  structure, exports and structures at Win 6+ level yet, right? 
Nope. But a
compatibility layer in usermode should be enough for now

>   How big is the danger that newer MSVC versions
> can't be used to 
>  compile ReactOS anymore? (E.g. loosing the capability to build Win
>  XP compatible files) 
We use a barebones compiler, without any SDK or something. The only
danger I see - they may remove support for old architectures or SEH
mechanism (CxxFrameHandler3). So far latest version work

And we always have GCC and clang, the latter is going to have a full
SEH support in some near future


Victor

On Thu, 2020-11-26 at 10:07 +0100, michael(a)fritscher.net wrote:
>   Moin,
>  
>  a few questions:
>     * How much work needs to be done to get newer versions of MSVC
>  running 
>  by e.g. using the shims? (not GUI, only the toolchain, and possibly
>  hack 
>  around installer problems)?
>     * How big is the danger to lose the capability to exchange ReactOS
>  and 
>  Win 2k3 system files?
>     * There isn't a time plan to get at least the core system file 
>  structure, exports and structures at Win 6+ level yet, right?
>     * How big is the danger that newer MSVC versions can't be used to 
>  compile ReactOS anymore? (E.g. loosing the capability to build Win
>  XP 
>  compatible files)
>  
>  I think the first question/solution could be a short time
>  compromise, 
>  the rest points at the "real" long time solution: targeting a newer 
>  Windows version..
>  
>  Best regards,
>  Michael Fritscher
>  
>  _______________________________________________
>  Ros-dev mailing list
>  Ros-dev(a)reactos.org
>  http://reactos.org/mailman/listinfo/ros-dev

[Mark Jansen]

Agreed with Colin.


On 26-Nov-20 09:27, Colin Finck wrote:
>   Am 25.11.2020 um 22:26 schrieb Hermès BÉLUSCA-MAÏTO:
> >   What about that discussion about why we have
> > ditched out MSVC 2010 
> >  support 
>  You mean reintroducing an ancient compiler that forces us to hack 
>  around its incapacity to deal with anything more modern than C89? One 
>  that already makes it harder to import third-party components and 
>  attract new developers?
>
>  Let's face it, we already make our incredible job of writing a 
>  Windows-compatible operating system even harder than Microsoft's - by 
>  supporting both GCC and MSVC. We spend enormous amounts of time to 
>  just get the basics right for both toolchains, and still fail at 
>  things like the C++ standard library. Ditching MSVC 2010 finally paves 
>  the way for importing a modern and trusted third-party C++ library 
>  like Clang's libc++ to replace our hacked and unmaintained stlport. 
>  Going even further, we could finally draw from the huge pool of 
>  talented C++ developers, who leverage the benefits of C++11 and 
>  beyond. Think about that next time you debug a memory leak or a 
>  deadlock that would have been impossible through std::unique_ptr or 
>  std::mutex.
>
>  This project won't ever become usable in a lifetime if we don't go 
>  with the time but stick to old habits forever. There is no inherent 
>  value in supporting old compilers, just wasted developer time.
>  And don't tell me about self-hosting: It's not like anybody 
>  productively uses ReactOS for building ReactOS right now. If you just 
>  want to demonstrate that, use the RosBE GCC-based toolchain. I already 
>  spent hours on backporting that back to ReactOS/Windows XP...
>
>
>  Now I usually would have put that on the agenda for the meeting and 
>  asked for a vote.
>  But as this year's ditching of MSVC 2010 is just the implementation of 
>  a meeting decision from 2 years ago 
>  (https://reactos.org/project-news/december-2018-meeting-minutes/), 
>  meeting votes are apparently not considered binding here. Neither is 
>  apparently the discussion we had in the related PR 
>  https://github.com/reactos/reactos/pull/2658.
>
>
>  So let's have an open discussion and voting process right here on the 
>  mailing list for everyone to see, to decide this once and for all!
>  Just reply to this mail, tell about your reasons - and more 
>  importantly tell whether you support or reject dropping MSVC 2010 
>  support.
>
>  Votes by CORE REACTOS MEMBERS are then counted on 1ST JANUARY 2021.
>
>
>  At this point, it should be obvious that I myself SUPPORT the decision 
>  to ditch building ReactOS with MSVC 2010.
>
>
>  Cheers,
>
>  Colin
>
>
>  _______________________________________________
>  Ros-dev mailing list
>  Ros-dev(a)reactos.org
>  http://reactos.org/mailman/listinfo/ros-dev

[Stas'M]

I think I agree with Colin on that.
 
I personally remember some people who said something like «meh… C89». You can try to
change my mind though, if you can answer two questions:
*  How can we attract new experienced developers who prefer new C standards while still
conforming to C89?
*  How can we make 3rd party modules maintenance to have less overhead while still having
support for MSVC2010?
​​​​​​​​​​​​​​Seriously regarding self-hosting I think we should focus on NT6+ application
compatibility shims to support newer MS CL compilers.
  
>  Thursday, 26 November 2020, 11:28 +03:00 from Colin
> Finck <colin(a)reactos.org>rg>:
>  
> Am 25.11.2020 um 22:26 schrieb Hermès BÉLUSCA-MAÏTO:
> >   What about that discussion about why we have
> > ditched out MSVC 2010 support 
> You mean reintroducing an ancient compiler that forces
> us to hack around
> its incapacity to deal with anything more modern than C89? One that
> already makes it harder to import third-party components and attract new
> developers?
>
> Let's face it, we already make our incredible job of writing a
> Windows-compatible operating system even harder than Microsoft's - by
> supporting both GCC and MSVC. We spend enormous amounts of time to just
> get the basics right for both toolchains, and still fail at things like
> the C++ standard library. Ditching MSVC 2010 finally paves the way for
> importing a modern and trusted third-party C++ library like Clang's
> libc++ to replace our hacked and unmaintained stlport. Going even
> further, we could finally draw from the huge pool of talented C++
> developers, who leverage the benefits of C++11 and beyond. Think about
> that next time you debug a memory leak or a deadlock that would have
> been impossible through std::unique_ptr or std::mutex.
>
> This project won't ever become usable in a lifetime if we don't go with
> the time but stick to old habits forever. There is no inherent value in
> supporting old compilers, just wasted developer time.
> And don't tell me about self-hosting: It's not like anybody productively
> uses ReactOS for building ReactOS right now. If you just want to
> demonstrate that, use the RosBE GCC-based toolchain. I already spent
> hours on backporting that back to ReactOS/Windows XP...
>
>
> Now I usually would have put that on the agenda for the meeting and
> asked for a vote.
> But as this year's ditching of MSVC 2010 is just the implementation of a
> meeting decision from 2 years ago
> ( https://reactos.org/project-news/december-2018-meeting-minutes/ ),
> meeting votes are apparently not considered binding here. Neither is
> apparently the discussion we had in the related PR
> https://github.com/reactos/reactos/pull/2658 .
>
>
> So let's have an open discussion and voting process right here on the
> mailing list for everyone to see, to decide this once and for all!
> Just reply to this mail, tell about your reasons - and more importantly
> tell whether you support or reject dropping MSVC 2010 support.
>
> Votes by CORE REACTOS MEMBERS are then counted on 1ST JANUARY 2021.
>
>
> At this point, it should be obvious that I myself SUPPORT the decision
> to ditch building ReactOS with MSVC 2010.
>
>
> Cheers,
>
> Colin
>
>  
> _______________________________________________
> Ros-dev mailing list
> Ros-dev(a)reactos.org
> http://reactos.org/mailman/listinfo/ros-dev
>    
 
 
--
Stanislav Motylkov

[Victor Perevertkin]

Many words were said already about dropping the support for old
compilers. I agree with all of them and want to put one more argument
on top of it from myself.

There are a lot of innovations happen in compilers and tooling. Big
companies put thousands of man-hours to improve compiler toolchains and
standard libraries (LLVM and libc++ are the headliners here -
everything goes there first).
It's silly not to use all of those, it makes finding bugs easier by
orders of magnitude (modern C++ makes you write code with less bugs
too, but that has been said already).
For example, AddressSanitizer will show you out of bounds access right
where it happened, with stack trace and memory contents.

But there is one thing here, it doesn't work without the standard
library support. Of course, we can write that into our own lib
(stlport?), reinventing the wheel. But isn't it better to spend time on
something ReactOS-related, where there is no open source code available
at all? (note: I'm asking about who is going to do that)

And if we choose to incorporate libc++ into our project, which supports
all of that tooling out of the box for Windows, we have another problem
- it can't be compiled with ancient compilers.

So basically, the choice is simple - either we support building ReactOS
with old compilers (this is not only about MSVC2010, but for everything
which doesn't support modern-enough C++), or we benefit from modern
tech for dynamic and static code analysis. I prefer the latter.


Regards,
Victor

[Hermès BÉLUSCA-MAÏTO]

And it's not possible to use these tools and libraries as plug-in and -out?
I mean, for example, for people interested in these, and some of our
buildbots, to do "configure+build using libc++ and LLVM" in order to get
more info about the code they write in ROS,
and in such a way that it is also possible for other people who want to
build ROS with e.g. MSVC 2010, to instead "configure+build using MSVC 2010 +
older lib" ? 

Regards,
Hermes

-----Original Message-----
From: Ros-dev <ros-dev-bounces(a)reactos.org> On Behalf Of Victor Perevertkin
Sent: 18. prosinca 2020. 05:13
To: ReactOS Development List <ros-dev(a)reactos.org>
Subject: Re: [ros-dev] Status Meeting (November 2020)

Many words were said already about dropping the support for old compilers. I
agree with all of them and want to put one more argument on top of it from
myself.

There are a lot of innovations happen in compilers and tooling. Big
companies put thousands of man-hours to improve compiler toolchains and
standard libraries (LLVM and libc++ are the headliners here - everything
goes there first).
It's silly not to use all of those, it makes finding bugs easier by orders
of magnitude (modern C++ makes you write code with less bugs too, but that
has been said already).
For example, AddressSanitizer will show you out of bounds access right where
it happened, with stack trace and memory contents.

But there is one thing here, it doesn't work without the standard library
support. Of course, we can write that into our own lib (stlport?),
reinventing the wheel. But isn't it better to spend time on something
ReactOS-related, where there is no open source code available at all? (note:
I'm asking about who is going to do that)

And if we choose to incorporate libc++ into our project, which supports all
of that tooling out of the box for Windows, we have another problem
- it can't be compiled with ancient compilers.

So basically, the choice is simple - either we support building ReactOS with
old compilers (this is not only about MSVC2010, but for everything which
doesn't support modern-enough C++), or we benefit from modern tech for
dynamic and static code analysis. I prefer the latter.


Regards,
Victor






_______________________________________________
Ros-dev mailing list
Ros-dev(a)reactos.org
http://reactos.org/mailman/listinfo/ros-dev

[Victor Perevertkin]

As I said, "It doesn't work without the standard library support".
You will need to use a different standard library for old compilers

Victor

On Mon, 2020-12-21 at 18:47 +0100, Hermès BÉLUSCA-MAÏTO wrote:
>   But there is one thing here, it doesn't work
> without the standard
>  library
>  
>  support.

[Colin Finck]

Hello everybody and a happy new year!

Am 26.11.2020 um 09:27 schrieb Colin Finck:
>   So let's have an open discussion and voting
> process right here on the 
>  mailing list for everyone to see, to decide this once and for all!
>  Just reply to this mail, tell about your reasons - and more importantly 
>  tell whether you support or reject dropping MSVC 2010 support.
>  
>  Votes by CORE REACTOS MEMBERS are then counted on 1ST JANUARY 2021. 
1st January has passed and it's time to count the votes cast on the 
mailing list.
First of all, a big thank you to the people who have submitted their 
opinions! Your votes matter to get the best possible overview on this topic.

It's a pity that many developers have not replied, when I'm sure that 
they have an opinion here.
On the other hand, it's quite understandable considering that this is 
already the 3rd (and definitely last!) time we are debating this matter.

So without further ado, here are the votes everyone can gather from the 
replies on ros-dev:


Dropping MSVC 2010 support for building ReactOS
===============================================
* Colin Finck
* Dominik Hornung
* Ged Murphy
* Mark Jansen
* Stanislav Motylkov
* Victor Perevertkin

Keeping MSVC 2010 support for building ReactOS
==============================================
* Doug Lyons
* Hermès Bélusca-Maïto
* Joachim Henze
* Oleg Dubinskij


Third discussion, third time it's the same result.
Thanks for providing a clear answer on that!

Note that I haven't applied the "CORE REACTOS MEMBERS" filter here, but 
that would lead to roughly the same number of votes getting subtracted 
on both sides.
I also leave it up to you to merge this result with the votes cast in 
https://github.com/reactos/reactos/pull/2658. However, there is just no 
calculation where the "Keeping" fraction has a majority.


I'm glad to say that this result allows us to continue improving 
ReactOS. By not sticking with abandoned compilers like MSVC 2010 
indefinitely, we can leverage the powers of C99 and C++11 and all 
third-party projects that have been developed after 2010.

Victor's recent integration of WDF/KMDF has already paved the way for a 
new storage stack that many of you could already test.
And an import of libc++ would finally fix the mess that our current C++ 
STL support is.


Cheers,

Colin

[Aleksey Bragin]

I won't waste your time by repeating what's already written here about 
dropping VS2010 support. Count my voice in for dropping the support for 
MSVC 2010.

Even though I love old computers, I always envisioned ReactOS as a 
modern operating system for modern computers. There are other 
communities for those who like old stuff (MS-DOS 6.22, Windows 3.11, 
Watcom compilers, etc).

If ReactOS can benefit from newest technologies made accessible by other 
- it should!


Regards,
Aleksey Bragin

On 1/7/2021 9:39 AM, Colin Finck wrote:
>   Hello everybody and a happy new year!
>
>  Am 26.11.2020 um 09:27 schrieb Colin Finck:
> >   So let's have an open discussion and voting
> > process right here on the 
> >  mailing list for everyone to see, to decide this once and for all!
> >  Just reply to this mail, tell about your reasons - and more 
> >  importantly tell whether you support or reject dropping MSVC 2010 
> >  support.
> >
> >  Votes by CORE REACTOS MEMBERS are then counted on 1ST JANUARY 2021. 
>  1st January has passed and it's time to count the votes cast on the 
>  mailing list.
>  First of all, a big thank you to the people who have submitted their 
>  opinions! Your votes matter to get the best possible overview on this 
>  topic.
>
>  It's a pity that many developers have not replied, when I'm sure that 
>  they have an opinion here.
>  On the other hand, it's quite understandable considering that this is 
>  already the 3rd (and definitely last!) time we are debating this matter.
>
>  So without further ado, here are the votes everyone can gather from 
>  the replies on ros-dev:
>
>
>  Dropping MSVC 2010 support for building ReactOS
>  ===============================================
>  * Colin Finck
>  * Dominik Hornung
>  * Ged Murphy
>  * Mark Jansen
>  * Stanislav Motylkov
>  * Victor Perevertkin
>
>  Keeping MSVC 2010 support for building ReactOS
>  ==============================================
>  * Doug Lyons
>  * Hermès Bélusca-Maïto
>  * Joachim Henze
>  * Oleg Dubinskij
>
>
>  Third discussion, third time it's the same result.
>  Thanks for providing a clear answer on that!
>
>  Note that I haven't applied the "CORE REACTOS MEMBERS" filter here, 
>  but that would lead to roughly the same number of votes getting 
>  subtracted on both sides.
>  I also leave it up to you to merge this result with the votes cast in 
>  https://github.com/reactos/reactos/pull/2658. However, there is just 
>  no calculation where the "Keeping" fraction has a majority.
>
>
>  I'm glad to say that this result allows us to continue improving 
>  ReactOS. By not sticking with abandoned compilers like MSVC 2010 
>  indefinitely, we can leverage the powers of C99 and C++11 and all 
>  third-party projects that have been developed after 2010.
>
>  Victor's recent integration of WDF/KMDF has already paved the way for 
>  a new storage stack that many of you could already test.
>  And an import of libc++ would finally fix the mess that our current 
>  C++ STL support is.
>
>
>  Cheers,
>
>  Colin