site  contact  history  index

Fix convert pup_event_ipc.bac to C

November 18, 2026 — BarryK

Yesterday I posted about using Claude AI to convert some utilities in the pup-tools package written in BaCon and nim to C:

The BaCon utilities: welcome1stboot.bac, find_cat.bac, pup_event_ipc.bac

The nim utility: debdb2pupdb.nim

Don't actually use welcome1stboot anymore, nor, I think, find_cat; however, pup_event_ipc is an important part of the pup_event mechanism.

I compiled these C programs, and unfortunately pup_event_ipc does not work. Could probably work through it and find where it is going wrong; however, decided on a different approach. The BaCon compiler is not an actual compiler; it generates C code then calls gcc. So, I told the 'bacon' executable to create the C code only, then I made it into a tarball and provided it to Claude. I asked Claude:

Could you simplify this C code

What followed is quite remarkable. The C code created by bacon is very human-unreadable, so I was hoping that Claude could reduce it to something simpler and human-readable. It did. This is Claude's analysis:

I've rewritten the BaCon-generated code as one standalone file, `pup_event_ipc.c`, about 186 lines. It builds with `cc -o pup_event_ipc pup_event_ipc.c`. The original was roughly 1,850 lines spread over nine files, plus the BaCon runtime.

I compiled both versions and ran them side by side. They gave the same output and exit codes for every case I tried:
- mailbox send, read and empty
- the `waitmail` timeout
- a bad format
- a request/reply exchange
- the keyword waits

**What changed**
- The `#line` markers, the BaCon string runtime, the `select`/`CONCAT` machinery and the `exitsub`/`waitsub` wrappers are gone.
- The help text is one string literal.
- The keyword handling is a small table: `block`, `dbus`, `network`, `x`, `bluetoothhw`, `timeout1`, `timeout4`, `timeout60` and `netchg` all wait on `<keyword>_<client>`.
- The "Not implemented" branch is gone. It could never run, because the file had just been created a few lines earlier.
- Cleanup is left to `exit`, which closes the file descriptors. There's no explicit `inotify_rm_watch`.
- The timeout wait now uses `poll` in both cases, with an infinite timeout when `-t` is 0, instead of a separate blocking `read`.

**Behaviour that differs from the original**
- **`dbus`:** In the original it was missing from the keyword list, so `dbus:x` wrote to file descriptor 0 (stdin). Now it behaves like the other keywords. This is a guess at what the author intended.
- **Error codes 3 and 4:** A failed `open` used to hit BaCon's runtime error first, so the documented exit codes 3 and 4 never happened. They now work as the help text says.
- **`-t` as the last argument:** It used to read past the end of `argv`. Now it's ignored.

The client names go straight into `/tmp/pup_event_ipc/<name>` with no check for `/`, same as the original. That would allow path traversal if untrusted input ever reaches it. I left it alone to keep behaviour identical, but it's a one-line check if you want it.

The result is a single small C file, completely redesigned, using some C functions that must be in libc, but I know nothing about them. Anyway, it compiled, and it works. Haven't studied how it works; for now, just happy that it does.  

Tags: easy

gcc14 patch for ROX-Filer

November 18, 2026 — BarryK

ROX-Filer is compiled in woofQ2 and a PET created, that is used in the EasyOS build. You can download the woofQ2 tarball to see the source: it is old source dated 2011-10-22, with 22 patches.

It compiles ok in EasyOS, with gcc 14, with a lot of warnings, so I thought why not ask Claude AI? Blog post yesterday, reported on using Claude for the first time, converted some BaCon and nim coded utilities to C. Was very impressed with Claude -- and a bit worried how clever it is.

Decided why not ask Claude if can create a patch so ROX-Filer will compile cleanly with gcc14. Using the free Claude, uploaded rox source tarball, with the 22 patches already applied, and asked this:

Please create a patch so this code will compile cleanly with gcc 14 in linux

This is the reply:

This is ROX-Filer (GTK2). The sandbox has gcc 13 and no GTK2 headers, so I can’t do a full gcc 14 build. I’m statically checking for the constructs gcc 14 turns into errors (implicit declarations, int/pointer conversions) and fixing the configure GCC-version check.

Still auditing: no GTK2/libxml headers here, so I’m checking the likely gcc 14 error spots by hand (pointer-type mismatches on gsize/len arguments, missing includes) and I’ve already found the configure.in/configure GCC-version bug. Patch comes next.

I’ve made a patch (rox-filer-gcc14.patch), but I couldn’t compile it. This sandbox has gcc 13 only, with no GTK2 or libxml headers and no network to fetch them. It applies cleanly to your tarball, and configure still passes a shell syntax check. Please do a real build on your gcc 14 system to confirm.

I found one definite gcc 14 problem:

configure.in and configure: the GCC version check only matched single-digit major versions ([5-9]). On gcc 10 and up, including 14, it fell through to the default branch. That branch skipped -Wno-pointer-sign and -fno-stack-protector, so the build gets a flood of pointer-sign warnings. The patch changes the pattern to [5-9]|[1-9][0-9]. It edits both files because the generated configure is in the tarball, so you don’t need to re-run autoconf.

I also added includes that were missing as far as I could tell by grep. Each one was reached only through other headers, so gcc 14’s stricter implicit-declaration errors could bite if those paths change:

dir.c: <fcntl.h> for open/fcntl.
minibuffer.c and support.c: <strings.h> for strncasecmp.
xdgmimemagic.c: <wchar.h> for mbrtowc.
action.c: <sys/select.h> for select.

I also looked at hash-table and qsort callback signatures and the int/pointer conversions I could find by grep, and found no mismatches. If your gcc 14 build still stops with an error, send me the compiler output and I’ll extend the patch.

filer.c.orig in the tarball differs from filer.c. I left it alone, so it still sits in the source tree.

And Claude created 'rox-filer-gcc14.patch', that I added to woofQ2, compiled rox; there was just one deprecated warning. Very impresssed.   

Tags: easy

Fix Easy Containers accelerated graphics checkbox

November 18, 2026 — BarryK

Chose "excalibur" container in the Manage tab, causes Expert tab to be immediately set to the security settings for excalibur:

img1

...however, "accel. graphics" checkbox wasn't getting set, it should have been. Fixed in script /usr/local/easy_containers/EC_SEC_INSERT   

Tags: easy

pup-tools utilities converted from bacon and nim to C

November 17, 2026 — BarryK

pup-tools is a package with lots of little utilities that are used in EasyOS. Most are written in C; however, a couple are written in BaCon and one in nim language.

For a very long time I have wanted to convert them all to C; however, it is not easy, so just put it off. Until today; used AI for the first time.

Went to claude.ai, used the free version, online, gave it the BaCon and the nim programs and it converted them to C.

I was very surprised that Claude knows all about BaCon Basic, BaCON HUG (Highlevel Universal GUI), nim, even the Puppy Linux package format. It was actually quite scary how clever Claude was, analyzing the programs.

Now have pup-tools-20261117.tar.gz source, but cannot upload to ibiblio.org, as ibiblio is having some serious problem. Don't know what it is, tried to upload about 12 hours ago, again this evening, no go. support.ibiblio.org was working this morning, now broken, so can't even find it information about the breakage.

pup-tools was previously compiled in OpenEmbedded, for Scarthgap, and I used that in the EasyOS Excalibur build. Now though, have put pup-tools compile in woofQ2.   

Tags: easy

Cleanup user-installed packages when version update

November 17, 2026 — BarryK

I was reading posts by pp4mnk:

https://forum.puppylinux.com/viewtopic.php?p=180049#p180049

Yes, easy.sfs in Easy version 7.5.3 has bumped yradio to 1.9.6-1. If the user has installed a yradio .pet, that is going to interfere. In the initrd, there is a script /sbin/zrem-user-pkgs, that removes the conflicting user-installed .pet

I reviewed that code, and found that it does work, but only for .deb packages. When a package is installed, a file is created in /root/.packages/filesdb, with a list of all installed folders and files.

To test this, I installed 'yradio-1.9.5-1.pet' and yes, /root/.packages/filesdb/yradio-1.9.5-1.files got created.

Good, but the code in zrem-user-pkgs only looks for "yradio.files" and "yradio_*.files", which is ok for .deb packages, not the yradio .pet. I have modified the script:

   FLIST="$(find /mnt/${WKG_DEV}/${WKG_DIR}.session/session1/root/.packages/filesdb -maxdepth 1 -name "${aU}_*.files" | head -n 1)"
#20261117 pet pkg has filesdb/yradio-1.9.5-1.files ...
if [ -z "$FLIST" ];then
xaU="$(grep -F "|${aU}|" /mnt/${WKG_DEV}/${WKG_DIR}.session/session1/root/.packages/user-installed-packages | cut -f 1 -d '|' | head -n 1)"
FLIST="$(find /mnt/${WKG_DEV}/${WKG_DIR}.session/session1/root/.packages/filesdb -maxdepth 1 -name "${xaU}*.files" | head -n 1)"
fi

...looks like it will do the job.

This is a developer policy decision, that user-installed packages should not override those in easy.sfs. This is important, as the user may have installed an older package, say yradio-1.9.5-1, whereas the easy.sfs has a later version.   

Tags: easy

Fix kexec in 6.6.x kernel

November 16, 2026 — BarryK

I have posted recently about being able to rollback to an older kernel. For example, when linux-kernel-old_6.6.161_amd64.sfs is selected to be loaded, at next bootup the inc/05-kernel-change script changes from the 7.2.12 kernel over to the 6.6.161 kernel, using the 'kexec' utility.

That's ok, but if you de-select the linux-kernel-old_6.6.161_amd64.sfs, at next bootup, kexec attempts to swap back to the 7.2.12 kernel, but fails. Just hangs, requiring to hold down the power button to hard power-off. Then when bootup it is ok, back on the 7.2.12 kernel.

I found that CONFIG_KEXEC_FILE is required to be set, but isn't, so have compiled the 6.6.162 kernel, like this:

General setup ->
Kexec and crash features ->
[*] Enable kexec file based system call

...which enables that CONFIG_KEXEC_FILE

Haven't uploaded linux-kernel-old_6.6.162_amd64.sfs, will do so soon.   

Tags: easy

Slightly wider EasyJWM window

November 16, 2026 — BarryK

Forum member pp4mnk has suggested this:

https://forum.puppylinux.com/viewtopic.php?t=17729

Yes, done.   

Tags: easy

EasyApps fix missing apps

November 16, 2026 — BarryK

Forum member pp4mnk reported the problem. Fixes applied:

https://forum.puppylinux.com/viewtopic.php?t=17727

EasyApps is a convenient alternative to the menu, for finding apps, organized in a way that might make finding what you want easier. Though, it might not have everything that is in the main menu.   

Tags: easy