Qlot can't install from git on Windows

[djha-skin]

Describe the bug

Reproducible steps

1. Get on a windows 11 box
2. Get scoop
3. From a prompt, run scoop install roswell
4. From the same prompt, run ros install sbcl-bin
5. From the same prompt, run ros install djhaskin987/qlot (I am preparing a windows pull request against master, but ros install fukamachi/qlot might work here as well.
6. Create a directory somewhere and add it to the windows PATH, like C:\Users\<you>\Executables.
7. Add this script to that folder under the name rsc.bat:

@echo off
ros -Q -L sbcl-bin -- C:\Users\bhw\.roswell\lisp\quicklisp\bin\%*

8. Check out the code at [email protected]:~skin/cl-i
9. Run rsc qlot init
10. In that folder, put this in the qlfile:

git com.djhaskin.nrdl [email protected]:~skin/nrdl :ref 0.4.0

12. Run rsc qlot install
13. Run rsc qlot exec ros run
14. Run (asdf:load-system "com.djhaskin.nrdl") in the REPL.

The REPL says the system is still not installed.

Error messages

After the console prints ● [1/1] com.djhaskin.nrdl Writing metadata files., nothing happens. The process returns with exit code -1073741819, and quits silently without continuing to check the repo out. I can tell by running (asdf:load-system "com.djhaskin.nrdl") in a REPL (via rsc qlot exec ros run) and it won't come up.

I have used print statements to narrow the problem down to the releases.txt function in src/utils/distify.lisp. For some reason, after the ironclad functions run, I can't get print statements to show up past those function calls.

This is odd, as I can copy and paste the (multiple-value-bind ...) form from that function into my REPL and it works fine. The ironclad functions work from the repl, just not when calling via qlot.

Versions:

• Qlot version: master branch
• Where did you get it from: roswell
• OS: Windows 11
• Lisp: SBCL 2.4.9

[djha-skin]

I continue to debug this error.

The code always dies at this line in src/utils/distify.lisp:

(ironclad:digest-file :md5 tarball-file)

This article mentions that

The exit code -1073741819 in hex is 0xC0000005. If you look up this code in Microsoft's documentation (NTSTATUS Values) you will see that it indicates that your program was terminated due to an access violation. This error can occur for a variety of reasons, including de-referencing a NULL pointer or referencing an invalid address.

You should step through your code in the debugger line by line to determine where the access violation happens. That should enable you to analyze the situation and identify the cause.

I also have continued to note that when run from my REPL, there is no error. I am starting to believe there is a threading issue. Some piece of code may be touching that tarball file when this code touches it as well, causing an access violation.

[djha-skin]

Reading further. Access violation is essentially a core dump/segfault. The system is segfaulting because somewhere an attempt to read a null pointer is happening. Again, this does not happen from the REPL alone.

[djha-skin]

I was correct -- it was the parallel processing. I disabled it in a pull request and it started working again. Please accept the PR (#291), it should fix things :)

Thanks for all your hard work, this tool is a joy to use.