librelist archives

Re: [shoes] Policeman packages Raisins exe. You decide

From:

Devyn Cairns

Date:

2010-03-22 @ 05:54

On Sun, Mar 21, 2010 at 9:11 PM, Cecil Coupe <ccoupe@cableone.net> wrote:

> Just another Ubuntu config problem. My ssh dot file have gone from linux
> to Os X to Linux and maybe back again - messy. I've got much to learn
> about git and github though.
>

If I understand you correctly, just create new keys for each machine. That's
what I've done.


>
> I have successfully committed my changes to my fork and my fork appears
> to be linked properly to the main fork. I added a list_box to the
> pack.rb dialog to chose the version of shoes to package with. You can't
> package a script for policeman if there is no policeman to download.
> Fixed up some nasty constant scoping issues 1.8.7 vs 1.9.1 (I think,
> thanks ashbb) I also committed the changes to binject.c
>
> The packer.rb changes, I'm not thrilled about but I see them as
> necessary and less Ruby ugly than my previous attempts and what was
> there. Less ugly. There is still work required in the /static/stub files
> to remove the hackeynet.
>

Ah. Somewhat related, is Ruby - or something similar - your primary
language? Either way, any kind of work is appreciated.


>
> I can't do the Windows work or the Os X work and they will require
> changes and then changes to pack.rb if Policeman (or the website)
> supports multiple versions.
>

That's okay. ashbb seems to be managing Windows fairly well :) and OS X 10.6
stuff is just getting started, but it shouldn't be too difficult. At the
moment, (although I may be wrong) the Linux version appears to be the most
stable, so that's what we should work toward achieving for OS X and Windows.
OS X and Linux share a lot of common ground, so there's not too much there.
Mostly Windows, because it's so different from the others; but as I said,
ashbb is doing a phenomenal job.


>
>
> On Sat, 2010-03-20 at 23:14 -0700, Devyn Cairns wrote:
> > really? I haven't had any ssh issues... ever... odd.
> >
> > On Sat, Mar 20, 2010 at 5:52 PM, Cecil Coupe <ccoupe@cableone.net>
> > wrote:
> >         On Fri, 2010-03-19 at 23:29 -0700, Devyn Cairns wrote:
> >         > If you could please fork shoes (if you haven't done so
> >         already) and
> >         > push your changes there? It's a lot easier to pull things in
> >         that way,
> >         > because then that commit is attributed to you.
> >         >
> >
> >
> >         Github and I are having some ssh issues before I can fork. I
> >         hate ssh
> >         'issues' so it might be long time.
> >
> >
> >
> >
> >
> > --
> >    ~devyn
>
>
>


-- 
   ~devyn

On Sun, 2010-03-21 at 22:54 -0700, Devyn Cairns wrote:
> On Sun, Mar 21, 2010 at 9:11 PM, Cecil Coupe <ccoupe@cableone.net>
> wrote:
>         Just another Ubuntu config problem. My ssh dot file have gone
>         from linux
>         to Os X to Linux and maybe back again - messy. I've got much
>         to learn
>         about git and github though.
> 
> 
> If I understand you correctly, just create new keys for each machine.
> That's what I've done.

Yeah that was the chosen solution. And then I had put the new public key
on all the other servers I ssh too. An annoyance. 
>  
>         
>         I have successfully committed my changes to my fork and my
>         fork appears
>         to be linked properly to the main fork. I added a list_box to
>         the
>         pack.rb dialog to chose the version of shoes to package with.
>         You can't
>         package a script for policeman if there is no policeman to
>         download.
>         Fixed up some nasty constant scoping issues 1.8.7 vs 1.9.1 (I
>         think,
>         thanks ashbb) I also committed the changes to binject.c
>         
>         The packer.rb changes, I'm not thrilled about but I see them
>         as
>         necessary and less Ruby ugly than my previous attempts and
>         what was
>         there. Less ugly. There is still work required in
>         the /static/stub files
>         to remove the hackeynet.
> 
> 
> Ah. Somewhat related, is Ruby - or something similar - your primary
> language? Either way, any kind of work is appreciated.
Devyn,

  I go way back to assembler languages back before the intel/datapoint
8008 was born. Cobol. Fortran. PDP-11, PDP-10, 8008, 6800, 6502,
8086/88, 68000/20/30. Pascal. C. Tiny bit of Lisp and Scheme. C++, Java.
Tcl, Python and Ruby. Ive been cross platorm  GUI for a long time. XVT
anyone? Star Division before Sun bought them. WxWindows, Java and then I
faded away, thinking that cross platform GUI will never happen. Then I
found Shoe (cute) but packaging was always the big problem to solve and
_why solved it. For a while, it a few minutes of web time, all was good.

Each language and framework has it's idioms and each has it's own
politically correct style.  I try to stay within the community rules as
I understand them. I'm a sensitive guy. T
>  
>         
>         I can't do the Windows work or the Os X work and they will
>         require
>         changes and then changes to pack.rb if Policeman (or the
>         website)
>         supports multiple versions.
> 
> 
> That's okay. ashbb seems to be managing Windows fairly well :) and OS
> X 10.6 stuff is just getting started, but it shouldn't be too
> difficult. At the moment, (although I may be wrong) the Linux version
> appears to be the most stable, so that's what we should work toward
> achieving for OS X and Windows. OS X and Linux share a lot of common
> ground, so there's not too much there. Mostly Windows, because it's so
> different from the others; but as I said, ashbb is doing a phenomenal
> job.
I don't *know* that Linux is any more stable. Back when I was running
Shoes and OS X, Linux had some issues that OS X and Windows didn't.
Linux may be easier to fix first. Or maybe not.

On Mon, Mar 22, 2010 at 12:32 AM, Cecil Coupe <ccoupe@cableone.net> wrote:

> On Sun, 2010-03-21 at 22:54 -0700, Devyn Cairns wrote:
> > On Sun, Mar 21, 2010 at 9:11 PM, Cecil Coupe <ccoupe@cableone.net>
> > wrote:
> >         Just another Ubuntu config problem. My ssh dot file have gone
> >         from linux
> >         to Os X to Linux and maybe back again - messy. I've got much
> >         to learn
> >         about git and github though.
> >
> >
> > If I understand you correctly, just create new keys for each machine.
> > That's what I've done.
>
> Yeah that was the chosen solution. And then I had put the new public key
> on all the other servers I ssh too. An annoyance.
> >
> >
> >         I have successfully committed my changes to my fork and my
> >         fork appears
> >         to be linked properly to the main fork. I added a list_box to
> >         the
> >         pack.rb dialog to chose the version of shoes to package with.
> >         You can't
> >         package a script for policeman if there is no policeman to
> >         download.
> >         Fixed up some nasty constant scoping issues 1.8.7 vs 1.9.1 (I
> >         think,
> >         thanks ashbb) I also committed the changes to binject.c
> >
> >         The packer.rb changes, I'm not thrilled about but I see them
> >         as
> >         necessary and less Ruby ugly than my previous attempts and
> >         what was
> >         there. Less ugly. There is still work required in
> >         the /static/stub files
> >         to remove the hackeynet.
> >
> >
> > Ah. Somewhat related, is Ruby - or something similar - your primary
> > language? Either way, any kind of work is appreciated.
> Devyn,
>
>  I go way back to assembler languages back before the intel/datapoint
> 8008 was born. Cobol. Fortran. PDP-11, PDP-10, 8008, 6800, 6502,
> 8086/88, 68000/20/30. Pascal. C. Tiny bit of Lisp and Scheme. C++, Java.
> Tcl, Python and Ruby. Ive been cross platorm  GUI for a long time. XVT
> anyone? Star Division before Sun bought them. WxWindows, Java and then I
> faded away, thinking that cross platform GUI will never happen. Then I
> found Shoe (cute) but packaging was always the big problem to solve and
> _why solved it. For a while, it a few minutes of web time, all was good.
>

Good to know. It may do you some good to really sit down and experiment with
a good dynamic language, obviously, for Shoes, that would be Ruby, and do a
bunch of uniquely-dynamic things, like complex metaprogramming etc. Then,
you'll really understand *why* that particular style is politically correct
in most cases. You don't have to, just a suggestion :D

I'm quite different, on the other hand, I started very high level, and only
recently am I starting to work my way down. I'm beginning to get more
comfortable with C and variants.

Granted, I'm also 13, so it was inevitable that I would start with a high
level programming language.


>
> Each language and framework has it's idioms and each has it's own
> politically correct style.  I try to stay within the community rules as
> I understand them. I'm a sensitive guy. T
>

That's usually a sensible thing to do... one of my friends claims to work
with Ruby a lot, but he doesn't even fully understand classes, because all
he writes are short little scripts that, most of the time, could just as
easily be written in bash ._.


> >
> >
> >         I can't do the Windows work or the Os X work and they will
> >         require
> >         changes and then changes to pack.rb if Policeman (or the
> >         website)
> >         supports multiple versions.
> >
> >
> > That's okay. ashbb seems to be managing Windows fairly well :) and OS
> > X 10.6 stuff is just getting started, but it shouldn't be too
> > difficult. At the moment, (although I may be wrong) the Linux version
> > appears to be the most stable, so that's what we should work toward
> > achieving for OS X and Windows. OS X and Linux share a lot of common
> > ground, so there's not too much there. Mostly Windows, because it's so
> > different from the others; but as I said, ashbb is doing a phenomenal
> > job.
> I don't *know* that Linux is any more stable. Back when I was running
> Shoes and OS X, Linux had some issues that OS X and Windows didn't.
> Linux may be easier to fix first. Or maybe not.
>
>
>
all I know is that when I was running Arch Linux & Shoes, compiled from
source a few months ago, the only thing that had issues was the packager,
otherwise, it ran perfectly... if I ever had a segfault, it would have been
because I was doing something stupid.



-- 
   ~devyn

> 
> Good to know. It may do you some good to really sit down and
> experiment with a good dynamic language, obviously, for Shoes, that
> would be Ruby, and do a bunch of uniquely-dynamic things, like complex
> metaprogramming etc. Then, you'll really understand why that
> particular style is politically correct in most cases. You don't have
> to, just a suggestion :D

We're getting OT for the list. I don't program professionally anymore,
but over the decades I've seen a lot of promising ideas and
technologies. Some worked out well but most didn't survive.  Ruby is an
interesting language. Shoes is an interesting project to work on. 

Yes it is. Now, back on topic :-)

On Mon, Mar 22, 2010 at 6:49 PM, Cecil Coupe <ccoupe@cableone.net> wrote:

>
> >
> > Good to know. It may do you some good to really sit down and
> > experiment with a good dynamic language, obviously, for Shoes, that
> > would be Ruby, and do a bunch of uniquely-dynamic things, like complex
> > metaprogramming etc. Then, you'll really understand why that
> > particular style is politically correct in most cases. You don't have
> > to, just a suggestion :D
>
> We're getting OT for the list. I don't program professionally anymore,
> but over the decades I've seen a lot of promising ideas and
> technologies. Some worked out well but most didn't survive.  Ruby is an
> interesting language. Shoes is an interesting project to work on.
>
>
>
>


-- 
   ~devyn

Hi Cecil, Debyn and folks,

Thank you for the encouragement. :-)

But about `Binject::EXE.new`,
I have still the same status as this post (Jan 13 9:28pm):
http://groups.google.com/group/shoooes/browse_thread/thread/d2e03c8f660790c1?pli=1

I can't execute the following snippet with my latest Policeman (0.r1380) for
Windows.
http://www.rin-shun.com/shoes/index.html

require 'binject'
Shoes.app do
  Binject::EXE.new(File.join(DIR, "static", "stubs", "blank.exe"))
  para 'hello'
end

> I also committed the changes to binject.c
Oh Cecil, did you modify `binject.c` ?
If so, could you show the code?

I also tried your `pack.rb`:
http://www.mvmanila.com/public/cecil-pack.rb

But, umm.... have no luck so far.

ashbb

Ashbb,

  I've become git-enabled. My changes are visible as a fork to the
master and  I've rewritten some of that posted code that is failing for
you.

  Yes, I did modify binject.c - three changes to replace rb_file_open()
with the older (previously working) rb_fopen(). A pain in the ass to
track down but the old code worked just fine - 3 args vs 2. 

  Try the lastest commit to pack.rb. I'm not saying it's great code but
it does attempt to address some of the issues you raised in the google
groups thread. 

Regards,
--Cecil

  

On Mon, 2010-03-22 at 15:35 +0900, Satoshi Asakawa wrote:
> Hi Cecil, Debyn and folks,
> 
> Thank you for the encouragement. :-)
> 
> But about `Binject::EXE.new`, 
> I have still the same status as this post (Jan 13 9:28pm):
> 
http://groups.google.com/group/shoooes/browse_thread/thread/d2e03c8f660790c1?pli=1
> 
> I can't execute the following snippet with my latest Policeman
> (0.r1380) for Windows.
> http://www.rin-shun.com/shoes/index.html
> 
> require 'binject'
> Shoes.app do
>   Binject::EXE.new(File.join(DIR, "static", "stubs", "blank.exe"))
>   para 'hello'
> end
> 
> > I also committed the changes to binject.c
> Oh Cecil, did you modify `binject.c` ?
> If so, could you show the code?
> 
> I also tried your `pack.rb`:
> http://www.mvmanila.com/public/cecil-pack.rb
> 
> But, umm.... have no luck so far.
> 
> ashbb
> 

On Mon, Mar 22, 2010 at 12:11 AM, Cecil Coupe <ccoupe@cableone.net> wrote:

> Ashbb,
>
>  I've become git-enabled.


Oh my god, it's become too powerful! =D


>  My changes are visible as a fork to the
> master and  I've rewritten some of that posted code that is failing for
> you.


>  Yes, I did modify binject.c - three changes to replace rb_file_open()
> with the older (previously working) rb_fopen(). A pain in the ass to
> track down but the old code worked just fine - 3 args vs 2.
>
>  Try the lastest commit to pack.rb. I'm not saying it's great code but
> it does attempt to address some of the issues you raised in the google
> groups thread.
>
> Regards,
> --Cecil


=D

-- 
   ~devyn

On Mon, 2010-03-22 at 00:24 -0700, Devyn Cairns wrote:
> On Mon, Mar 22, 2010 at 12:11 AM, Cecil Coupe <ccoupe@cableone.net>
> wrote:
>         Ashbb,
>         
>          I've become git-enabled.
> 
> Oh my god, it's become too powerful! =D

Damn freaking metaphors. 

Hi Cecil,

Thank you for uploading your shoes fork on github.
I looked at your commit:
http://github.com/ccoupe/shoes/commit/6560000e9cd4ca7355c3f38c41a8229dd9fb8f86

I modified binject.c as same as yours and tried to compile with MinGW on
Windows XP.
But, ... had no luck, couldn't compile. :(

Is it possible to use rb_fopen() with Ruby 1.9.1?

Umm... I must have misunderstood something, though...

ashbb

ashbb,

  There is definitely something wrong.  Looking in the system's 1.9.1
headers, there is no rb_fopen.  in 1.8.7 there is. There is an
rb_file_open in both with the proper arguments in both. 

  So, why did it fail for me using 1.8.7 rb_file_open? Raisins'
binject.c used rb_fopen so that's what I reverted to. Perhaps the 1.8.7
implementation of rb_file_open doesn't match the header? 

  Until I figure it out, we should conditionally compile those three
lines depending of the ruby version.

--Cecil

On Wed, 2010-03-24 at 00:31 +0900, Satoshi Asakawa wrote:
> Hi Cecil,
> 
> Thank you for uploading your shoes fork on github.
> I looked at your commit:
> http://github.com/ccoupe/shoes/commit/6560000e9cd4ca7355c3f38c41a8229dd9fb8f86
> 
> I modified binject.c as same as yours and tried to compile with MinGW
> on Windows XP.
> But, ... had no luck, couldn't compile. :(
> 
> Is it possible to use rb_fopen() with Ruby 1.9.1?
> 
> Umm... I must have misunderstood something, though...
> 
> ashbb
> 

Hi Cecil,

Thank you for the great research. Now I understand my issue for Windows.

- I can complile Binject with Ruby 1.9.1 mingw32.
- But this snippet chrashes (runtime error)

    require 'binject'
    Shoes.app do
      Binject::EXE.new 'blank.exe'
    end

- Crash point is line 531 in binject.c

http://github.com/ashbb/shoes/blob/master/req/binject/ext/binject_c/binject.c#L531

BINJ_READ macro is just one line. I've not find a solution so far.
But I think I'll be able to find something... hopefully. ;-)

Cheers,
ashbb

There's lots of compiler messages with binject and ruby 1.9.1 which
can be fixed by including the proper headers I believe. BTW, you can can
cd into req/binject/ext/binject_c and run 'make' to recompile any mods
to binject.c. When the error messages are just the 'fread return not
used' then you can go make to top-level and do the longer rake and
build.

On Wed, 2010-03-24 at 22:50 +0900, Satoshi Asakawa wrote:
> Hi Cecil,
> 
> Thank you for the great research. Now I understand my issue for
> Windows.
> 
> - I can complile Binject with Ruby 1.9.1 mingw32.
> - But this snippet chrashes (runtime error)
> 
>     require 'binject'
>     Shoes.app do
>       Binject::EXE.new 'blank.exe'
>     end 
> 
> - Crash point is line 531 in binject.c
> 
> 
http://github.com/ashbb/shoes/blob/master/req/binject/ext/binject_c/binject.c#L531
> 
> BINJ_READ macro is just one line. I've not find a solution so far.
> But I think I'll be able to find something... hopefully. ;-)
> 
> Cheers,
> ashbb

Definitely something wrong in binject.c/rb_file_open() and ruby 1,8.7 vs
1.9.1

I've add #ifdef's to use rb_fopen for 1.8.7 and rb_file_open for 1.9.1
at:
http://github.com/ccoupe/shoes/commit/777094a2b0482754d1542d1931e3e294ba08d436

However, it fails on 1.9.1 for me.  Might work you. It could be a
patch_level issue with my Ruby 1.9.1 versus the lastest stable 1.9.1.
Theres a lot of notes in the Ruby 1.9.1 Changelog about rb_file_open. 


On Tue, 2010-03-23 at 13:02 -0600, Cecil Coupe wrote:
> ashbb,
> 
>   There is definitely something wrong.  Looking in the system's 1.9.1
> headers, there is no rb_fopen.  in 1.8.7 there is. There is an
> rb_file_open in both with the proper arguments in both. 
> 
>   So, why did it fail for me using 1.8.7 rb_file_open? Raisins'
> binject.c used rb_fopen so that's what I reverted to. Perhaps the 1.8.7
> implementation of rb_file_open doesn't match the header? 
> 
>   Until I figure it out, we should conditionally compile those three
> lines depending of the ruby version.
> 
> --Cecil

Apologies if this is too long but I need some opinions and it does
matter to getting Shoes Policeman running in both 1.8.7 and 1.9.1

Actually the binject problem is much worse than I thought for 1.9.1. A
lot of those extensions and gems in req/ spit out a lot of warning
messages during compile on Ubuntu Ruby 1.9.1. Warnings that indicate the
#defines and #includes are not right for 1.9.1. Even though it compiles
and links, functions can be called that do not exist in the libraries.
Seqfault Ick!

For binject.c it would be nice to have the Ruby version passed in on the
compile command line (-DRUBY_1_9) as is done for most of the compile
tasks in the Rakefile. Which leads to  the binject_c/extconf.rb which
generates the makefile to be used to compile .c files in that directory

I don't have a warm fuzzy feeling about all this mkmf stuff. A hacker
would do something like the Rakefile does everywhere else: add the
-DRUBY_1_9 to the CFLAGS based on the ruby being used and then fix all
the code (binject.c) to include the proper .h files based on the
RUBY_1_9 define.

A possibly better idea would be to add RUBY_1_8 or RUBY_1_9 (one or the
other) to the mkmk.rb CFLAGS, and then the maintainers of the
binject_c/* could #ifdef their code appropriately. Or someone could
figure out how to pass the Ruby-version from the Rakefile to the
extconf.rb files

Like I said, I'm not comfortable in this part of the build process. Your
opinion is desired. Compiling Sqlite3 in 1.9.1 may need a similar fix to
it's mkmf.rb
--Cecil

Cecil and all,

apologies, not had time to look at your email in detail, but here's a
suggestion for releasing Policeman:

Why don't we drop support for building exes, dmgs and run files and just
make sure that Shy files are working?

We could just remove the bits from the packaging UI that relate to building
for OSX, Win and Linux, we don't have to remove all the code.

Then, since it seems sorting packaging out is going to take some time, we
could work on that for a 3.5 release?

Thoughts?


-----------------------
i5m.co.uk
GPG Key: 0xA18A602B


On Wed, Mar 24, 2010 at 3:46 AM, Cecil Coupe <ccoupe@cableone.net> wrote:

> Apologies if this is too long but I need some opinions and it does
> matter to getting Shoes Policeman running in both 1.8.7 and 1.9.1
>
>
>

On Wed, 2010-03-24 at 12:58 +0000, i5m wrote:
> Cecil and all,
> 

> 
> Why don't we drop support for building exes, dmgs and run files and
> just make sure that Shy files are working?

Do we know shy files aren't working? I don't use then myself and I don't
see any shy issues reported at the git mothership.
> 
> 
> We could just remove the bits from the packaging UI that relate to
> building for OSX, Win and Linux, we don't have to remove all the code.
Actually it would be easy to modify pack.rb to conditionally
include/exclude the gui checkboxs when running in ruby 1.9.1 

No need to hide full functionality for 1.8.7 users where things work
fine. 
> 
> 
> Then, since it seems sorting packaging out is going to take some time,
> we could work on that for a 3.5 release?
> 
> 
Actually, someone will probably figure out the 1.9.1 issues before the
snow leopard port arrives.  I need to take a few days off to file my
taxes.

Cecil,

On Thu, Mar 25, 2010 at 12:51 AM, Cecil Coupe <ccoupe@cableone.net> wrote:

> On Wed, 2010-03-24 at 12:58 +0000, i5m wrote:
>
> Do we know shy files aren't working?
>

I phrased that badly. I'm pretty sure shy files are working, but what I
meant is that we should make sure creating them still works ok in Policeman.
For instance, on OSX PPC the main packaging window won't open (I get a crash
when I try to open it), but the simple packager that Mental implement does
work ok.

>
> No need to hide full functionality for 1.8.7 users where things work
> fine.


Ok, but I think that Shoes will only build with 1.8.7 for Linux users. We've
probably broken things on the Windows and OSX front and it will require
1.9.



> >
> > >
> Actually, someone will probably figure out the 1.9.1 issues before the
> snow leopard port arrives.  I need to take a few days off to file my
> taxes.
>

Well, we can live in hope.

Minor status report.

Binject.c is just C source code that includes a maze of include files
with Macros. Turns out ruby 1.9.1 changed the internal definition of a
file struct and the access macros into the struct and the names of the
macros. Added a layer or two of structs->*structs and changed some of
the Macro sugar that hides it. With that kind of code changed you'd
think binject.c would just fail to compile at all. 

It would have been better for us if it did. I know where the problem is.
Solving won't be that hard except I've got RSI and can't use the
keyboard to too long.

On Wed, 2010-03-24 at 18:51 -0600, Cecil Coupe wrote:
> On Wed, 2010-03-24 at 12:58 +0000, i5m wrote:
> > Cecil and all,
> > 
> > 
> > Why don't we drop support for building exes, dmgs and run files and
> > just make sure that Shy files are working?
> 
> Do we know shy files aren't working? I don't use then myself and I don't
> see any shy issues reported at the git mothership.
> > 
> > 
> > We could just remove the bits from the packaging UI that relate to
> > building for OSX, Win and Linux, we don't have to remove all the code.
> Actually it would be easy to modify pack.rb to conditionally
> include/exclude the gui checkboxs when running in ruby 1.9.1 
> 
> No need to hide full functionality for 1.8.7 users where things work
> fine. 
> > 
> > 
> > Then, since it seems sorting packaging out is going to take some time,
> > we could work on that for a 3.5 release?
> > 
> > 
> Actually, someone will probably figure out the 1.9.1 issues before the
> snow leopard port arrives.  I need to take a few days off to file my
> taxes.
> 
> 

Hi Cecil et al,

Let me post a status report for building Policeman Windows.

Now trying to build `binject`:
- Marged Cecil's work and customized for my building environment, Windows
MinGW. (sorry, not add ifdef so far)
- Compile is ok now, I can make `binject.so` file from scratch.
- It can run and create .exe file.
- But the .exe file doesn't work well.
- I guess the cause is `binject_exe_rewrite` function in binject.c

Look at the commits: http://github.com/ashbb/shoes/commits/master

Now in progress, but I feel I had fewer crashes. So, I think it's ok to
release for the first Policeman. ;-)

ashbb

ashbb,

  It's worse than you think with binject.c 
The code is written to assume a 'C' FILE structure ptr is returned from
the old rb_fopen call which it stuffs away in the BINJ struct. 

In ruby 1.9.1,  rb_file_open returns a VALUE (an RFILE) which has to be
dereferenced to get an rb_io_t struct*. There's no FILE structure in
that (ignore the stdio_file one). They switched to using file
descriptors (an int) in 1.9.1 instead of a FILE *. so fread(),fwrite(),
etc,  will compile (grrr) but hang, segfault, or half way work
randomly. 

It's not that hard to replace fread() with read() but tedious.
I'm experimenting with just using a simple fopen() call instead of the
rb_file_open calls. Sort of works but it also fails. This issue requires
a serious rethink about what binject does, how it does it and the
completely undocumented ruby 1.9.1 embedding API and how to test it. 

Tonight, I think it's too fragile to fix. Who ever does fix it will own
it forever and I don't want to be that person. I don't want to own 'C'
code. I might feel differently tomorrow if an elegant solution magically
appears.

--Cecil

On Sat, 2010-03-27 at 17:26 +0900, Satoshi Asakawa wrote:
> Hi Cecil et al,
> 
> Let me post a status report for building Policeman Windows.
> 
> Now trying to build `binject`:
> - Marged Cecil's work and customized for my building environment,
> Windows MinGW. (sorry, not add ifdef so far)
> - Compile is ok now, I can make `binject.so` file from scratch.
> - It can run and create .exe file.
> - But the .exe file doesn't work well.
> - I guess the cause is `binject_exe_rewrite` function in binject.c
> 
> Look at the commits: http://github.com/ashbb/shoes/commits/master
> 
> Now in progress, but I feel I had fewer crashes. So, I think it's ok
> to release for the first Policeman. ;-)
> 
> ashbb

Hi Cecil,

Thank you so much for the detail explanation.
You are doing awesome work!

I didn't know what I was doing. :-P
Now I understand we are faced with a tough problem.

I have no experience to solve such a tough problem.
It may be more than I can handle.

Umm... hope some experts help us.
I'll do my best, but step by step. ;-)

ashbb

I managed to get binject and ruby 1.9.1 working for .exe in my latest
commit and sorted out 
http://github.com/ccoupe/shoes/commit/7e0ac87ee760695e72a8f4f9bfb67618ff114114

It is not well tested. It hangs building dmg which is odd since my
changes shouldn't have touched that code. There are plenty of troubling
warnings from gcc when compiling with the 1.9.1 headers, so the task is
far from complete.

It would be handy if embedded ruby API experts out there take a look at
the code. I'm worried about garbage collection and/or memleaks -
although if your packaging a script that probably isn't a big deal.

On Sun, 2010-03-28 at 22:51 +0900, Satoshi Asakawa wrote:
> Hi Cecil,
> 
> Thank you so much for the detail explanation.
> You are doing awesome work!
> 
> I didn't know what I was doing. :-P
> Now I understand we are faced with a tough problem.
> 
> I have no experience to solve such a tough problem.
> It may be more than I can handle.
> 
> Umm... hope some experts help us.
> I'll do my best, but step by step. ;-)
> 
> ashbb

I finally found some thinking about why building a dmg in 1.9.1 can
hang. Not fail or seg fault. Just hang. Copying a few files from here
than there should be fast. Perhaps the threading changes in 1.9.1 are
the issue? lib/shoes/pack.rb does set up a thread (for the progress bar)
where all the binject stuff happens and the binject.c code calls back
into the shoes/ruby code block. Going from green threads to native
threads could cause a hang (deadlock)

Maybe one of you 1.9.1 thread savvy could take a look at that code and
pronounce it thread worthy for 1.9.1. Actually, does the progress bar
really work in 1.9.1 on a multi-core machine with long running threads?
Remembering some of ashbb's code in his pack.rb, I think he might have
been struggling with a progress bar for downloading, perhaps the same
threading issue. Or not.

I am not a threading expert. I don't want to be one. 

Hi Cecil,

Yes. I met the same thread issue when I was checking simple-rubygems.rb with
my local Policeman for Windows.

In that case, I couldn't improve (debug) Shoes C code, so changed a little
bit Ruby code in setup.rb.

This is the commit:
http://github.com/ashbb/shoes/commit/bd749be8d5c553dd56600f46945c80f2786a5f7f

Hope this helps,
ashbb

... This is exactly what I was planning on doing for HacketyHack, so I vote
yay on that proposition.

+1

On Wed, Mar 24, 2010 at 6:00 AM, Steve Klabnik <steve@steveklabnik.com>wrote:

> ... This is exactly what I was planning on doing for HacketyHack, so I vote
> yay on that proposition.




-- 
   ~devyn

ashbb,

  It could be a difference between 1,8,7 and 1.9.1, although it works
for me in both. There is an (generated?) intern.h file which has the
other definition but I haven't looked into how that is created. It's
remotely possible the functions are different in Windows ruby than the
other runtimes. The rb_file_open function is only called from binject.c,
the prototype in intern.h, but NO source file defines it. 

Time to see how the ruby files are compiled. 

It's very confusing.


On Wed, 2010-03-24 at 00:31 +0900, Satoshi Asakawa wrote:
> Hi Cecil,
> 
> Thank you for uploading your shoes fork on github.
> I looked at your commit:
> http://github.com/ccoupe/shoes/commit/6560000e9cd4ca7355c3f38c41a8229dd9fb8f86
> 
> I modified binject.c as same as yours and tried to compile with MinGW
> on Windows XP.
> But, ... had no luck, couldn't compile. :(
> 
> Is it possible to use rb_fopen() with Ruby 1.9.1?
> 
> Umm... I must have misunderstood something, though...
> 
> ashbb
>