VirtualMachine vs. Platform and Bootstrap
Hi, 1. We can query for the VIRTUAL MACHINE and ask if it is 32 or 64 bit (internally depending on wordSize): Smalltalk vm is32Bit Anything OK on that side. 2. We can also query for the OPERATING SYSTEM we are running on using: Smalltalk os and there are subclasses for OSPlatform like Win32Platform, Win64Platform as well as Unix32Platform, unix64Platform, ... But when I evaluate "Smalltalk os" in Pharo 7 Build #70552 2a) I receive a Win32Platform instance which means 32 bit although I'm running on a Win 64 system 2b) I receive a Unix32Platform instance which means 32 bit although I'm running on a Ubuntu 64 system I see this as a clear bug and very confusing The reason for this behavior is simple: because I run with the 32 bit VM/image on these two 64 bit operating systems and in Pharo: for 2a) the#isActivePlatform method ask the VM (???)for the platform name (which is Win32 then) for 2b) the#isActivePlatform method also considers the word size from the VM (???) But to me the "os" should give me infos about the underlying operating system and is currently returning a wrong instance (because it falls back to querying about the running VM, not the real platform) This would be easy to fix with some UFFI calls (for instance piping from "uname -i" on Unix or "ver" on Windows). With UFFI I would also be able to give more and better infos in the platform hierarchy about the underlying OS. But doing this would introduce a dependency from the "Systems-Platforms" package to "UFFI" (and currently my understanding is that UFFI comes later)
From what I understood from the comment in getPwdViaFFI:with we also would like to focus on a minimal dependency for bootstrap ... and only depend on FFI and not UFFI.
Any comments how we could deal with that best? Thx T.
It's a question of responsibilities. Right now, it's used to determine which dll's to use on linux (on linux, see ffiModuleName), which means the distinction between 32/64 platforms is which ABI to use when doing UFFI, or more accurately, where to find libraries for the ABI supported by the vm. To make things less confusing, it would be better to remove the 32/64 subclasses from platform, (since they are really a separation only relevant to UFFI) and have the ffiModuleName: extension branch on Smalltalk vm is32/64 as appropriate for UnixPlatform. Then one would be free to implement either 32/64 subclasses as you suggest if there's prim support, or a #getDetailedPlatformInfo dependent on FFI to fetch more specific details like 32/64 bit, actual windows version, etc.. -- Sent from: http://forum.world.st/Pharo-Smalltalk-Developers-f1294837.html
On Fri, Feb 16, 2018 at 10:18 AM, Torsten Bergmann <astares@gmx.de> wrote:
Hi,
1. We can query for the VIRTUAL MACHINE and ask if it is 32 or 64 bit (internally depending on wordSize):
Smalltalk vm is32Bit
Anything OK on that side.
2. We can also query for the OPERATING SYSTEM we are running on using:
Smalltalk os
and there are subclasses for OSPlatform like Win32Platform, Win64Platform as well as Unix32Platform, unix64Platform, ...
But when I evaluate "Smalltalk os" in Pharo 7 Build #70552
2a) I receive a Win32Platform instance which means 32 bit although I'm running on a Win 64 system 2b) I receive a Unix32Platform instance which means 32 bit although I'm running on a Ubuntu 64 system
I see this as a clear bug and very confusing
Yup.
The reason for this behavior is simple: because I run with the 32 bit VM/image on these two 64 bit operating systems and in Pharo:
for 2a) the#isActivePlatform method ask the VM (???)for the platform name (which is Win32 then)
for 2b) the#isActivePlatform method also considers the word size from the VM (???)
But to me the "os" should give me infos about the underlying operating system and is currently returning a wrong instance (because it falls back to querying about the running VM, not the real platform)
This would be easy to fix with some UFFI calls (for instance piping from "uname -i" on Unix or "ver" on Windows). With UFFI I would also be able to give more and better infos in the platform hierarchy about the underlying OS.
But doing this would introduce a dependency from the "Systems-Platforms" package to "UFFI" (and currently my understanding is that UFFI comes later)
From what I understood from the comment in getPwdViaFFI:with we also would like to focus on a minimal dependency for bootstrap ... and only depend on FFI and not UFFI.
Well, but that's because the working directory is needed at bootstrap time. Take into account that the packages that belong to the bootstrap are those listed in BaselineOfPharoBootstrap System-Platforms is listed there, but maybe it is not so necessary? We should check what are the usages of it. In any case, what I'd say is that "stuff" that is not really needed at bootstrap time should be loaded afterwards, maybe as a separate package with extension methods? And in that case you're free to go and use uFFI, that should be no problem. Just take into account that uFFI is loaded by BaselineOfMorphicCore nowadays.
Any comments how we could deal with that best?
Thx T.
-- Guille Polito Research Engineer Centre de Recherche en Informatique, Signal et Automatique de Lille CRIStAL - UMR 9189 French National Center for Scientific Research - *http://www.cnrs.fr <http://www.cnrs.fr>* *Web:* *http://guillep.github.io* <http://guillep.github.io> *Phone: *+33 06 52 70 66 13
On 19 February 2018 at 17:35, Guillermo Polito <guillermopolito@gmail.com> wrote:
On Fri, Feb 16, 2018 at 10:18 AM, Torsten Bergmann <astares@gmx.de> wrote:
Hi,
1. We can query for the VIRTUAL MACHINE and ask if it is 32 or 64 bit (internally depending on wordSize):
Smalltalk vm is32Bit
Anything OK on that side.
2. We can also query for the OPERATING SYSTEM we are running on using:
Smalltalk os
and there are subclasses for OSPlatform like Win32Platform, Win64Platform as well as Unix32Platform, unix64Platform, ...
But when I evaluate "Smalltalk os" in Pharo 7 Build #70552
2a) I receive a Win32Platform instance which means 32 bit although I'm running on a Win 64 system 2b) I receive a Unix32Platform instance which means 32 bit although I'm running on a Ubuntu 64 system
I see this as a clear bug and very confusing
Yup.
The reason for this behavior is simple: because I run with the 32 bit VM/image on these two 64 bit operating systems and in Pharo:
for 2a) the#isActivePlatform method ask the VM (???)for the platform name (which is Win32 then)
for 2b) the#isActivePlatform method also considers the word size from the VM (???)
But to me the "os" should give me infos about the underlying operating system and is currently returning a wrong instance (because it falls back to querying about the running VM, not the real platform)
Good point, but to be pedantic I'd say "not the real OS" rather than "not the real platform". Would it be correct to distinguish between OS and Platform? where Platform is your compile target, and OS is where you run the exe ?? Thus the fix would not be just for "Smalltalk os" ==> Win64Platform but to introduce "Smalltalk platform" ==> Win64Platform for 64 bit VM and ==> Win32Platform for 32 bit VM. and "Smalltalk os" ==> something else entirely. or maybe I'm completely off base. cheers -ben
This would be easy to fix with some UFFI calls (for instance piping from "uname -i" on Unix or "ver" on Windows). With UFFI I would also be able to give more and better infos in the platform hierarchy about the underlying OS.
But doing this would introduce a dependency from the "Systems-Platforms" package to "UFFI" (and currently my understanding is that UFFI comes later)
From what I understood from the comment in getPwdViaFFI:with we also would like to focus on a minimal dependency for bootstrap ... and only depend on FFI and not UFFI.
Well, but that's because the working directory is needed at bootstrap time. Take into account that the packages that belong to the bootstrap are those listed in
BaselineOfPharoBootstrap
System-Platforms is listed there, but maybe it is not so necessary? We should check what are the usages of it. In any case, what I'd say is that "stuff" that is not really needed at bootstrap time should be loaded afterwards, maybe as a separate package with extension methods? And in that case you're free to go and use uFFI, that should be no problem.
Just take into account that uFFI is loaded by BaselineOfMorphicCore nowadays.
Any comments how we could deal with that best?
Thx T.
--
Guille Polito
Research Engineer
Centre de Recherche en Informatique, Signal et Automatique de Lille
CRIStAL - UMR 9189
French National Center for Scientific Research - *http://www.cnrs.fr <http://www.cnrs.fr>*
*Web:* *http://guillep.github.io* <http://guillep.github.io>
*Phone: *+33 06 52 70 66 13 <+33%206%2052%2070%2066%2013>
participants (4)
-
Ben Coman -
Guillermo Polito -
Henrik Sperre Johansen -
Torsten Bergmann