Ironclad 3 wip

[slozier]

Follow-up of discussion with @BCSharp in #16.

@BCSharp , in case you're interested, I pushed some more tweaks to my wip branch. I managed to get it running a tkinter hello world:

import os
os.environ["TCL_LIBRARY"] = r"C:\Python34\tcl\tcl8.6"

import ironclad
import sys
sys.path.append(r"C:\Python34\DLLs")

import tkinter as tk

root = tk.Tk() 
a = tk.Label(root, text ="Hello World") 
a.pack() 
  
root.mainloop() 

[BCSharp]

That's great! If you are OK with the commits I have pushed to ipy3-wip, I can rebase your commits on top of it. And if you are satisified with the so-far-code in your wip, then maybe merge it into ipy3-wip?

I have also a wip-on-wip branch named ipy3-clang on which I have experimental SCons code to build ic_mscvr90.dll using Clang, with PDB generation. Comments welcome. Once it proves itself (i.e. debugging works as expected), I intend to use Clang for python34.dll too.

The biggest challenges with Clang I've encountered:

1. lld-link seems to be unable to link directly with a DLL. I used the gendef/dlltool combo to generate the implib, which works OK, but still does not feel as good as what ld is capable of. Or maybe there is an lld option for that which I am not aware of.
2. ld exports all symbols from a DLL by default. lld-link "understands" the switch -export-all-symbols (i.e. doesn't bark at it as unrecognized), but still does not want to export all symbols automatically. This is not an issue for ic_msvcr90.dll, which is small and is to become even smaller, but for python34.dll a DEF will need to be generated during build. Or again, maybe there is a way to get lld to behave, but I simply don't know it yet.

[slozier]

Alright, I did a bit of cleanup and pushed to ipy3-wip. Unless you have any objections I'm thinking we might as well merge it to the main branch?

As for the ipy3-clang branch, assuming the debugging business all works then great! Better than the printf debugging I've been doing. 😄 My knowledge when it comes to C compilers is somewhat lacking so I'm probably not going to be of much help.

[BCSharp]

I think it's a good idea to merge it to the main branch. I have just a few questions and one request.

1. The commit getting hashlib (mostly) working introduced a regression: testAddObjectWithExistingReferenceAddsMappedObjectAndDecRefsPointer is failing. Are you aware of that?

2. About branch naming: you have mentioned merging it to the main branch. Did you mean the master branch? I have noticed that CPython, NumPy have renamed their 'master' to 'main' and new git repos default now with the name 'main' (which is arguably more readable). If you are considering following this convention, I think now would be a good moment to do the rename.

3. About Ironclad 2.7: although it was never fully working, I think it may still be useful to keep it somehow accessible, if not for reference only. How about renaming branch Docker (which has the last commit before ipy3 commits started, I believe) to '2.7'? I'd still update the Dockerfile on that branch to use the windows image with .NET preinstalled, to make the toolset easily accessible.

4. Going forward, what is your preferred way of working? Working with direct commits onto main or via pull requests? Perhaps something like this: trivial, or boring commits directly to 'main', more interesting changes via a PR, which, after sitting proposed for a day or two, if it doesn't attach any comments, can be merged (or squashed?) to main by the author. I don't want to create unneeded review pressure on either of us.

Request: let the merge to 'main' (or 'master') be a fast-forward, rather than a squash. I'd like to see the history of all the changes, there were plenty of them. Perhaps this is your preference too.

[slozier]

1. I don't quite remember, but I think some reference counts were going below zero so I probably disabled the decref. I'll try to take a look at it in the coming days.
2. & 3. Alright, I did the following renaming:

• ipy3-wip to main and made it the default branch. The master branch is now gone.
• docker to 2.7

4. I'm fine with commits to main unless you think it's something worth reviewing. Until the repo becomes more stable I'm not super worried about it. I guess I should look into setting up some github actions to run the test suite...

[slozier]

1. I reverted the change to get the test passing as I couldn't get it to fail with the code I was testing. But I assume I'll run into it again at some point. 😄

[BCSharp]

A suggestion: to change the About text on the GitHub project page to something like:

CPython API compatibility layer for IronPython

I got that from the original Google Code page for Ironclad and find more informative than what we have now.

[slozier]

Works for me. Done.

[BCSharp]

The project root directory contains license.txt, which is 14 years old and likely needs to be updated. At least to address the .NET Foundation. The original license seems to be the MIT license, but the standard .NET Foundation license is Apache (that is, as used by IronPython).

Just something to think about.

[slozier]

Indeed something to think about. I'm not sure the .NET Foundation aspect covers this project as it was set up specifically with IronPython and the DLR in mind. in any case, I'm fine with sticking to the MIT license for this.

[GeorgeS2019]

.NET Foundation prefers MIT license. If agree, please update so more people could join the contribution

[slozier]

@BCSharp Getting an access violation exception since 92006a when trying to load up _hashlib , any ideas?

[BCSharp]

I've fixed the assembly code. _hashlib should import now.

[BCSharp]

Strange that test_hashlib doesn't fail.

Mystery solved: I was inadvertently using the wrong test_hashlib (the one from Tests). Will use a different name for the Ironclad test going forward.

[GeorgeS2019]

I know those guys at Ikvm is also using clang to get jdk for.NET6+ working

[GeorgeS2019]

@BCSharp
Is there major block moving Ironclad forwards?

[BCSharp]

From my side — availability of time.

[GeorgeS2019]

should this project be jumpstarted in 2025?

is there alternative?