www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - Dwarf Exception Handling now on FreeBSD!

reply Walter Bright <newshound2 digitalmars.com> writes:
32/64 support now on Linux and FreeBSD.

   https://github.com/D-Programming-Language/dmd/pull/5376

Turns out that FreeBSD is close enough to Linux that it "just worked".

Recently, there was a long thread entitled "C++17" where people are asking for 
better C++ interoperability. Well, here's a ripe plum for anyone wanting 
valuable compiler street cred!

Make the Dwarf EH support work on OSX 32 and 64.

I bet that nearly all the work will be in changing the ELF fixups to MACH-O 
fixups in the gcc_except_table and eh_frame sections. Use an object dumper to 
see what g++ uses for fixups in those sections, and do the same.

As for me, I'll be implementing C++ multiple inheritance class layout, because 
Manu's problems have made it clear we have to do that.
Jan 30 2016
next sibling parent ZombineDev <valid_email he.re> writes:
On Sunday, 31 January 2016 at 06:34:26 UTC, Walter Bright wrote:

 32/64 support now on Linux and FreeBSD.

   https://github.com/D-Programming-Language/dmd/pull/5376

 Turns out that FreeBSD is close enough to Linux that it "just 
 worked".

 Recently, there was a long thread entitled "C++17" where people 
 are asking for better C++ interoperability. Well, here's a ripe 
 plum for anyone wanting valuable compiler street cred!

 Make the Dwarf EH support work on OSX 32 and 64.

 I bet that nearly all the work will be in changing the ELF 
 fixups to MACH-O fixups in the gcc_except_table and eh_frame 
 sections. Use an object dumper to see what g++ uses for fixups 
 in those sections, and do the same.

 As for me, I'll be implementing C++ multiple inheritance class 
 layout, because Manu's problems have made it clear we have to 
 do that.


Nice work!

I suggest that after that you look into RAII C++ interop (calling 
C++ destructors at the end of the scope), because it is critical 
for interfacing with modern C++ projects that make heavy use of 
it.

Adding dummy methods as the first members of an interface to 
satisfy the vtbl layout is quite error-prone and useless when you 
need to make the cleanup from the D side!
Jan 31 2016
prev sibling next sibling parent reply Charles <csmith.ku2013 gmail.com> writes:
On Sunday, 31 January 2016 at 06:34:26 UTC, Walter Bright wrote:

 Well, here's a ripe plum for anyone wanting valuable compiler 
 street cred!

 Make the Dwarf EH support work on OSX 32 and 64.



Are there any future plans for Win64 since it won't ever support 
dwarf exceptions from how I understand it? Or does it, and I'm 
strongly mistaken?
Jan 31 2016
parent reply Walter Bright <newshound2 digitalmars.com> writes:
On 1/31/2016 6:32 PM, Charles wrote:

 Are there any future plans for Win64 since it won't ever support dwarf
 exceptions from how I understand it? Or does it, and I'm strongly mistaken?


VC++'s Win64 exception handling scheme is not at all the Win32 scheme and 
definitely not Dwarf EH. Supporting it would be an additional level of effort
on 
par with supporting Dwarf EH, presuming documentation on it can be found. (I 
haven't looked for it.)

If someone wants to search for a spec for it, that would be a good start.
Jan 31 2016
parent reply Rikki Cattermole <alphaglosined gmail.com> writes:
On 01/02/16 4:01 PM, Walter Bright wrote:

 On 1/31/2016 6:32 PM, Charles wrote:

 Are there any future plans for Win64 since it won't ever support dwarf
 exceptions from how I understand it? Or does it, and I'm strongly
 mistaken?


 VC++'s Win64 exception handling scheme is not at all the Win32 scheme
 and definitely not Dwarf EH. Supporting it would be an additional level
 of effort on par with supporting Dwarf EH, presuming documentation on it
 can be found. (I haven't looked for it.)

 If someone wants to search for a spec for it, that would be a good start.


I believe the last link will be most helpful. But first will give you a 
good understanding and where to go(its for the linker).

If anybody wishes to look more into it.

https://msdn.microsoft.com/en-us/library/1deeycx5.aspx
https://msdn.microsoft.com/en-us/library/swezty51.aspx
https://msdn.microsoft.com/en-us/library/1eyas8tf.aspx
Jan 31 2016
parent reply Walter Bright <newshound2 digitalmars.com> writes:
On 1/31/2016 7:04 PM, Rikki Cattermole wrote:

 On 01/02/16 4:01 PM, Walter Bright wrote:

 On 1/31/2016 6:32 PM, Charles wrote:

 Are there any future plans for Win64 since it won't ever support dwarf
 exceptions from how I understand it? Or does it, and I'm strongly
 mistaken?


 VC++'s Win64 exception handling scheme is not at all the Win32 scheme
 and definitely not Dwarf EH. Supporting it would be an additional level
 of effort on par with supporting Dwarf EH, presuming documentation on it
 can be found. (I haven't looked for it.)

 If someone wants to search for a spec for it, that would be a good start.


 I believe the last link will be most helpful. But first will give you a good
 understanding and where to go(its for the linker).

 If anybody wishes to look more into it.

 https://msdn.microsoft.com/en-us/library/1deeycx5.aspx
 https://msdn.microsoft.com/en-us/library/swezty51.aspx
 https://msdn.microsoft.com/en-us/library/1eyas8tf.aspx


They have bits and pieces of the info, but nothing about what is actually 
generated to, say, catch an exception.
Jan 31 2016
next sibling parent Rikki Cattermole <alphaglosined gmail.com> writes:
On 01/02/16 4:23 PM, Walter Bright wrote:

 On 1/31/2016 7:04 PM, Rikki Cattermole wrote:

 On 01/02/16 4:01 PM, Walter Bright wrote:

 On 1/31/2016 6:32 PM, Charles wrote:

 Are there any future plans for Win64 since it won't ever support dwarf
 exceptions from how I understand it? Or does it, and I'm strongly
 mistaken?


 VC++'s Win64 exception handling scheme is not at all the Win32 scheme
 and definitely not Dwarf EH. Supporting it would be an additional level
 of effort on par with supporting Dwarf EH, presuming documentation on it
 can be found. (I haven't looked for it.)

 If someone wants to search for a spec for it, that would be a good
 start.


 I believe the last link will be most helpful. But first will give you
 a good
 understanding and where to go(its for the linker).

 If anybody wishes to look more into it.

 https://msdn.microsoft.com/en-us/library/1deeycx5.aspx
 https://msdn.microsoft.com/en-us/library/swezty51.aspx
 https://msdn.microsoft.com/en-us/library/1eyas8tf.aspx


 They have bits and pieces of the info, but nothing about what is
 actually generated to, say, catch an exception.


Hmm you're right.
But on the plus side I found this:
http://www.programmingunlimited.net/siteexec/content.cgi?page=libseh

It even supports Digital Mars C/C++ compiler!
Jan 31 2016
prev sibling parent Charles <csmith.ku2013 gmail.com> writes:
On Monday, 1 February 2016 at 03:23:18 UTC, Walter Bright wrote:

 They have bits and pieces of the info, but nothing about what 
 is actually generated to, say, catch an exception.


Gotcha, is this something you'd expect to be easily obtainable 
(e.g. ask Microsoft Rep saying we want to make it easier for 
people to use their product?), or is this something that would 
probably require some sort of reverse engineering effort?
Feb 01 2016
prev sibling parent reply Jonathan M Davis <jmdavisProg gmx.com> writes:
On Sunday, 31 January 2016 at 06:34:26 UTC, Walter Bright wrote:

 32/64 support now on Linux and FreeBSD.

   https://github.com/D-Programming-Language/dmd/pull/5376

 Turns out that FreeBSD is close enough to Linux that it "just 
 worked".


It's always nice when that happens, especially when there's not 
much FreeBSD-specific work going on, and the autotesters have 
generally been on older versions of FreeBSD such that issues on 
the newer versions don't always get noticed (e.g. for a while 
there, the shared library stuff passed the tests on FreeBSD 8 but 
not later; fortunately Martin got that fixed though).

- Jonathan M Davis
Feb 03 2016
parent reply Brad Roberts via Digitalmars-d <digitalmars-d puremagic.com> writes:
On 2/3/16 11:28 AM, Jonathan M Davis via Digitalmars-d wrote:

 On Sunday, 31 January 2016 at 06:34:26 UTC, Walter Bright wrote:

 32/64 support now on Linux and FreeBSD.

   https://github.com/D-Programming-Language/dmd/pull/5376

 Turns out that FreeBSD is close enough to Linux that it "just worked".


 It's always nice when that happens, especially when there's not much
FreeBSD-specific work going on,
 and the autotesters have generally been on older versions of FreeBSD such that
issues on the newer
 versions don't always get noticed (e.g. for a while there, the shared library
stuff passed the tests
 on FreeBSD 8 but not later; fortunately Martin got that fixed though).

 - Jonathan M Davis


I haven't put much time into investigation, but the last time I tried it,
neither 9 nor 10 passed 
the test suite.  If someone puts in the effort to get either or both of those
working, I'd be happy 
to upgrade some of the freebsd testers to newer versions.  There's currently 4
freebsd machines, so 
plenty of room to have a mix of versions.  Not enough to test every combination
of version and 
bitness, but enough to allow randomness to expose issues.
Feb 03 2016
next sibling parent reply Jacob Carlborg <doob me.com> writes:
On 2016-02-03 21:18, Brad Roberts via Digitalmars-d wrote:


 I haven't put much time into investigation, but the last time I tried
 it, neither 9 nor 10 passed the test suite.  If someone puts in the
 effort to get either or both of those working, I'd be happy to upgrade
 some of the freebsd testers to newer versions.  There's currently 4
 freebsd machines, so plenty of room to have a mix of versions.  Not
 enough to test every combination of version and bitness, but enough to
 allow randomness to expose issues.


Perhaps it's worth taking a look at Docker. As far as I understand 
there's experimental support for FreeBSD [1]. It uses ZFS, jail and the 
64bit Linux compatibility layer. Seems to be possible to run both Linux 
and FreeBSD images in Docker for FreeBSD.

Although I'm not sure if it's possible to mix 32bit and 64bit with Docker.

[1] https://wiki.freebsd.org/Docker

-- 
/Jacob Carlborg
Feb 04 2016
parent reply Brad Roberts via Digitalmars-d <digitalmars-d puremagic.com> writes:
On 2/4/2016 1:27 AM, Jacob Carlborg via Digitalmars-d wrote:

 On 2016-02-03 21:18, Brad Roberts via Digitalmars-d wrote:


 I haven't put much time into investigation, but the last time I tried
 it, neither 9 nor 10 passed the test suite.  If someone puts in the
 effort to get either or both of those working, I'd be happy to upgrade
 some of the freebsd testers to newer versions.  There's currently 4
 freebsd machines, so plenty of room to have a mix of versions.  Not
 enough to test every combination of version and bitness, but enough to
 allow randomness to expose issues.


 Perhaps it's worth taking a look at Docker. As far as I understand
 there's experimental support for FreeBSD [1]. It uses ZFS, jail and the
 64bit Linux compatibility layer. Seems to be possible to run both Linux
 and FreeBSD images in Docker for FreeBSD.

 Although I'm not sure if it's possible to mix 32bit and 64bit with Docker.

 [1] https://wiki.freebsd.org/Docker


It's not about mechanism, it's about compute hours in the day.  Yes, 
making each machine a little more flexible would be useful for leveling 
out the progress across platforms, but there just isn't enough hardware 
in my fleet for adding a bunch more configurations to the matrix.
Feb 04 2016
parent Jacob Carlborg <doob me.com> writes:
On 2016-02-04 10:43, Brad Roberts via Digitalmars-d wrote:


 It's not about mechanism, it's about compute hours in the day.  Yes,
 making each machine a little more flexible would be useful for leveling
 out the progress across platforms, but there just isn't enough hardware
 in my fleet for adding a bunch more configurations to the matrix.


Ah, I see. What's the average time for a pull request for the complete 
matrix?

-- 
/Jacob Carlborg
Feb 04 2016
prev sibling next sibling parent Joakim <dlang joakim.fea.st> writes:
On Wednesday, 3 February 2016 at 20:18:52 UTC, Brad Roberts wrote:

 On 2/3/16 11:28 AM, Jonathan M Davis via Digitalmars-d wrote:

 On Sunday, 31 January 2016 at 06:34:26 UTC, Walter Bright 
 wrote:

 [...]


 It's always nice when that happens, especially when there's 
 not much FreeBSD-specific work going on,
 and the autotesters have generally been on older versions of 
 FreeBSD such that issues on the newer
 versions don't always get noticed (e.g. for a while there, the 
 shared library stuff passed the tests
 on FreeBSD 8 but not later; fortunately Martin got that fixed 
 though).

 - Jonathan M Davis


 I haven't put much time into investigation, but the last time I 
 tried it, neither 9 nor 10 passed the test suite.  If someone 
 puts in the effort to get either or both of those working, I'd 
 be happy to upgrade some of the freebsd testers to newer 
 versions.  There's currently 4 freebsd machines, so plenty of 
 room to have a mix of versions.  Not enough to test every 
 combination of version and bitness, but enough to allow 
 randomness to expose issues.


One option to consider is installing FreeBSD 10 and then testing 
8 and 9 in jails, which are the precursor to Linux containers.  I 
used this approach years ago, when compiling Chromium for several 
FreeBSD versions.
Feb 04 2016
prev sibling parent reply Jonathan M Davis <jmdavisProg gmx.com> writes:
On Wednesday, 3 February 2016 at 20:18:52 UTC, Brad Roberts wrote:

 On 2/3/16 11:28 AM, Jonathan M Davis via Digitalmars-d wrote:

 On Sunday, 31 January 2016 at 06:34:26 UTC, Walter Bright 
 wrote:

 32/64 support now on Linux and FreeBSD.

   https://github.com/D-Programming-Language/dmd/pull/5376

 Turns out that FreeBSD is close enough to Linux that it "just 
 worked".


 It's always nice when that happens, especially when there's 
 not much FreeBSD-specific work going on,
 and the autotesters have generally been on older versions of 
 FreeBSD such that issues on the newer
 versions don't always get noticed (e.g. for a while there, the 
 shared library stuff passed the tests
 on FreeBSD 8 but not later; fortunately Martin got that fixed 
 though).

 - Jonathan M Davis


 I haven't put much time into investigation, but the last time I 
 tried it, neither 9 nor 10 passed the test suite.  If someone 
 puts in the effort to get either or both of those working, I'd 
 be happy to upgrade some of the freebsd testers to newer 
 versions.  There's currently 4 freebsd machines, so plenty of 
 room to have a mix of versions.  Not enough to test every 
 combination of version and bitness, but enough to allow 
 randomness to expose issues.


I know that the druntime and Phobos unit tests pass on 11 (which 
is what I'm currently running), and I'm fairly sure that they do 
on 10 (I don't know about 9), but I have no idea what the state 
of the dmd tests are on anything newer than 8. I'll have to see 
if I can find time to at least see what's currently working and 
report issues.

- Jonathan M Davis
Feb 06 2016
parent Walter Bright <newshound2 digitalmars.com> writes:
On 2/6/2016 6:15 AM, Jonathan M Davis wrote:

 I know that the druntime and Phobos unit tests pass on 11 (which is what I'm
 currently running), and I'm fairly sure that they do on 10 (I don't know about
 9), but I have no idea what the state of the dmd tests are on anything newer
 than 8. I'll have to see if I can find time to at least see what's currently
 working and report issues.

 - Jonathan M Davis


Thanks, Jonathan!
Feb 06 2016