Hi Doru, can you give us more insight of the errors? Because there should not be changes in the behavior of the VM. Maybe there are issues, we have used the Pharo tests as a guarantee and we are laking some tests. Cheers, Pablo. On Sun, Aug 11, 2019 at 9:04 PM Tudor Girba <tudor@tudorgirba.com> wrote:

Hi,

...

On Aug 11, 2019, at 5:29 PM, ducasse <stepharo@netcourrier.com> wrote:

...

On 11 Aug 2019, at 13:19, Tudor Girba <tudor@tudorgirba.com> wrote:

Excellent news!

I should say that for GT we are currently using the minheadlessVM (Ronie’s work) out of the opensmalltalk-vm repo (https://bintray.com/opensmalltalk/vm/cog) and it works remarkably well.

We started to play with your minheadlessVM. So far it looks like there are differences, although I am sure they are not large.

like what?

I do not know yet, but running GT in headless opens the window with the one from opensmalltalk-vm, but not with the new one. We did not yet look into details, but we will.

Cheers, Doru

...

...

So, what is the difference between the minheadlessVM built by you versus the one from opensmalltalk-vm?

Cheers, Doru

...

On Aug 8, 2019, at 9:53 AM, tesonep@gmail.com wrote:

TL;DR; ======

For the anxious, you can get real headless vm and image from zero-conf.

$ wget get.pharo.org/64/80+vmHeadlessLatest | bash

Zero conf scripts remain unchanged for users.

However, if you are launching the VM by hand from the executable instead of the launcher scripts (pharo and pharo-ui) as in

$ ./pharoexecutable Pharo.image

the image will launch in headless mode and will not open a window. To launch it in headfull, you can use the --interactive argument after the image, which will make the image open a window using SDL2.

$ ./pharoexecutable Pharo.image --interactive

Long version ============

Hi, this mail is the happy intermediate result of the work that us, the Pharo Consortium Team, has been doing in the last couple of months. Our main objective is to have a real headless implementation of Pharo where all the responsibility to open or not a World window (or other) is handled by the image. For doing so we have done a series of modifications in the image and the VM side. We consider this is the path that Pharo 8 and following versions should follow, as it will severely improve server-side and command line Pharo and in building custom desktop applications.

These modifications are available only in 64-bits machines (Windows, OSX, and Linux). ARM32 and 64bits headless is in the roadmap, but it is delayed because we have prioritized our three major platforms for this first couple of months.

All this work is based in Opensmalltalk-VM and Ronnie's initial work on headless. We are really grateful to all the contributors in the history of this nice product. To achieve a real headless VM we have brought modifications in the source tree because most of the platform code to open and manipulate windows is not required anymore. Instead, we use the SDL2 library that implements a nice layer on top of the OS and allows us to manage on the image side through FFI.

So this mail is now an open call for (beta?)testing. The sources of the current VM we are building are in the headless branch in https://github.com/pharo-project/opensmalltalk-vm And we have set up a CI that is both building and testing the VM in https://ci.inria.fr/pharo-ci-jenkins2/job/pharo-vm/job/headless/

For the future we have a lot of ideas, that will wait for another long email or a beer-talk @ESUG. We want to hear your ideas!!

Image-Side Improvements =======================

- The image handles the creation or not of the main world window. - We incorporated the idea of World renderer, where different backends are used to render the world. - We have 3 backends: VM support (compatibility with non-headless VMs), and OSWindow with two backends: SDL and GTK3+. - The modifications in the image are fully backward compatible with the non-headless VM and are pushed since weeks in the latest 8.0 image. - We move the handling of events to the image side when using SDL and GTK3+, opening the door to a richer set of events and finer-grained control over them. - SDL and GTK versions are implemented using FFI calls.

VM-Side Improvements ====================

- VMMaker code migrated to Tonel thanks to Feenk and included in the repository of the VM. - Making VMMaker execute in Pharo 7 and 8. - Removing GPL code from the VM repository (GDB).

- Slowly adding new tests for the JIT / Slang and VMGeneration. - Restructuring of the source code. - A new simpler CMake build. - Generate VM code from Slang on each build. - A CI process to validate (including the run of the tests in Pharo and the ones adding to the VM). - Simplification of the codebase.

- Maximize the reuse of code between the platforms (preferring the standard versions over the platform-specific). - Cleaning up duplicated code. - All the plugins are now external plugins. - The VM is now a dynamic library. This is a first step towards embedding Pharo into other applications. - The main executable is a thin frontend (you can change it or implement your own).

- Removing unused plugins. - Improved crash dump. Especially the crash dump works now in Windows 64bits. - Dummy implementation of Security plugin (it is going away.... eventually). - Cleanup of SSL, UUID, and Socket plugin.

- Cleanup of conditional code (Still to improve). - Improving the types used in the functions (we have to be neat to be multiplatform/multi-arch). - Improving the lookup of modules - Improving the logging of the VM - Improving the handling of VM arguments

Thanks a lot for reading so long!! We hope you enjoy the VM and please tell us all the problems you find!!

Pablo, Guille, and Esteban

-- feenk.com

"Being happy is a matter of choice."

-- feenk.com

"Speaking louder won't make the point worthier."

-- Pablo Tesone. tesonep@gmail.com