Hi, Andreas Enge <[email protected]> writes:
> Am Thu, May 28, 2026 at 11:06:22PM +0900 schrieb Maxim Cournoyer: >> "cannot be considered free software" is preposterous, no? We have >> doubts, but no certainty as of now. For our own source code, it seems >> wise to avoid the uncertainty creeping in, but for packages... I think >> given the gray area, and the low risk of it (we can easily remove such >> packages if/when it becomes clear that they aren't free software) > > This is a claim that I think is not correct. In a sense, we can only go > forward. Of course *theoretically*, we can take out any package; but > things become complicated as soon as dependents enter the game (and it > has taken a bit of effort to work on the question for GCD 006, and each > and every package removal means additional work). And also think of > updates - the vibe coding could happen between version X-1 and version X > of a package. It may be very difficult to downgrade (already our user > tools are not designed for this). > > Just imagine that we decided we wanted to roll back the Python 3.11 to > 3.12 transition - I think it is close to impossible. If this is a real concern, I think this pledge needs to add a special exception for high impact packages based on number of dependents. As far as I understand, the original proposal was not trying to cover core software that start to use LLM assistance. If it were the case that the next Python, Rust, Linux, etc versions start to use LLMs in a way that this pledge excludes them out of Guix, we should have some measure to not even need to discuss that we need this packages. As you say, it's not reasonable to drop, for example, Python. So, to avoid any subjectivity, we could add a clear numeric hard-line. Just as a clarification before the example, if Python would start to depend on some library which is fully vibe-coded, we will still consider it as a core software that needs to be packaged. I propose to use all package dependencies as a metric, not only direct. As a concrete proposal. Almost 86% of Guix packages depend on `python-wrapper`. I think we should do some analysis and set up a metric where we would do an exception. For example, we could add: ``` 3. Software where most of the code was authored or co-authored by genAI will not be packaged in Guix. Notable examples of such code include Claude’s C compiler, EmDash, and Neomacs. Exceptionally software that will have a high surface impact on the Guix package base, will be packaged; *without discussion*: 0.30% of the package base, around 100 dependents; *with case by case deliberation*: under that threshold; if the package does not have dependents 0% (eg: a new package), there is no need to discuss, this pledge excludes it from being included. ``` Note that Neomacs and Emdash would fall under the category of 0%, so they remain excluded. The amendment allows protecting, or at least deliberating, about what we will drop if it's already used by the community. For example, `orca` only has 2 dependencies. If it where to depend on a vibe-coded library, it will not be possible to update to newer versions since the dependency won't be packaged. In my opinion, no matter what, an accessibility software, such a screen reader, cannot be removed, it will be very hostile action for non-sighted users. Best regards, Sergio
