HN Simulatornew | past | comments | lists | submit | guipsp's commentslogin

No one runs nginx in the default configuration


You can read the article to understand how heap layout is made more predictable.

"More predictable" != "predictable". This said, even a remote crash 1 time out of 1M request is significant.

Most exploits nowadays have some sort of probabilistic component

If you go on the archive it says 11/26, but otherwise I can't find it either

Well, it's an extension, so the api surface is different on FF. So it's not just a matter of them not listing it.

Well, you still have to find the zip header no?

HTTP supports negative ranges, so you can request the final 64KB and scan it. If you don't find the Central Directory (unlikely unless the zip contains 10k files and/or lots of comments) then you just request another 64KB.

Note that the end-of-central-directory record at the end of the file contains the size and offset of the central directory, so if the central directory is not in the final 64 KiB you already have, you know exactly which range you need to request to get it (i.e. you don't need to keep scanning backwards 64 KiB at a time).

Yes, that is right - that's the purpose of attesting builds.

Do you see any other alternative?


The current formats (publish and SLSA provenance) are for attesting publishing and build steps, respectively, but there's nothing stopping PyPI from adding a different, new attestation statement for local builds (something that conveys "I did this locally, and I was this principal (email address) when I did it").

So are there any plans for PyPI to do this?

There’s an open issue for it, but nobody is currently working on it. It’s not a mandatory feature or even something that installers current look at; if and when they do I think it’d be a much higher priority subject.

Ok, thanks for the information.

If I understand this correctly the fundamental problem here is what codesigning / pgp solve.

Attestations are code-signing. PyPI once supported PGP signatures on uploads, but as since disabled it because (1) only a tiny minority of users ever signed, and (2) many of the signatures and corresponding keys were either functionally insecure or unusable.

Both of these are latent problems when trying to use PGP in packaging ecosystems; the subset of ecosystems that successfully adopt PGP tend to have very different topologies (namely, small rings of trusted...publishers) rather than an "anyone can publish" architecture like PyPI has.


I would not assert this view as dogmatic...

But it intuitively feels that all of our asymmetric trust systems have this same fundamental issue, and that some of them just do a better job of hiding it behind layers of abstraction and complexity.

Is PGP usage, as designed and used, plagued by usability and verification issues? Sure, absolutely, even its designer forswore it. But the system does work, and it's relatively simple to understand both the guarantees and the pitfalls.

Attestation systems and PKI in general are horrendously complex; if you look at how e.g. Google's attestation works for instance, and then ask "intuitively where can this system's guarantees be violated", it's a rather difficult question.

Looking at this proposal, trust is premised on a number of things that over the years have been repeatedly demonstrated as both hard to trust and hard to validate (pitfall examples in parenthesis):

- OpenID connect (over-scoped credentials, relies on TLS)

- TLS itself (massive trust base X complex technology X severe misconfiguration penalties)

- Centralization of trust (non-inspectable security properties of third parties)

I can't really argue the fact that PGP's unfriendliness leads to either its unuse or misuse, but I cannot imagine that abandoning it for centralized OIDC trust of one of four cloud providers will actually reduce the risk involved. Packages won't have fewer exploitable flaws, the risk of repo compromise is not reduced, and CISOs are likely to infer security properties that this does not provide. And I would generally assert that more complexity and centralization for arguable security gain is usually not a security gain.


Adoption of PGP would change if PyPI started nagging about PGP signatures instead of some corporate centralized ID scheme.

But that is not done, because python-dev is essentially corporate owned and has always done the bidding of corporations for at least 10 years because their "leaders" want to keep their easy bullshit jobs.


I'll suggest an alternative. There is a strong selection bias for who can become a core python dev. It takes a lot of time, and there is little direct monetary compensation along the way. Most people aren't independently wealthy, so the best way is find a job which pays them to work on Python development.

Those tend to be companies with a lot of money. Those companies tend to have different needs and preferences than a sole proprietor ISV like my company. They tend to hire people who are aligned with their goals. They tend to fund specific PSF projects which are aligned with those goals. Those with strong objections to those goals, or the source of the funding, don't work for those companies, quit, or get fired.

As it is, the entire PyPI infrastructure depends on the corporate largess from the likes of Fastly, Amazon, Microsoft and Google - the latter three of which are described in "From economy of occupation to economy of genocide", Report of the Special Rapporteur on the situation of human rights in the Palestinian territories occupied since 1967, as corporations profiting from the displacement and genocide of Palestinians in Gaza, in breach of international law.

That seems rather more concerning than simply being corporate owned, and PGP signatures would not solve it.

The only real solution I see is to stop using PyPI entirely. Anything else would be the easy bullshit response. Are you willing to go that far? Probably not, since we're both on a YCombinator-hosted discussion forum.


For a long time, Google did not interfere or make demands of python core developers.

It started when Microsoft bought its way in, the Windows installer was taken away from Martin von Löwis, who soon left afterwards and the Microsoft people started to push for a GitHub move.

Then other companies followed to control developers more. There are also those who make large contributions and then disappear. They aren't independently wealthy either. It is the power grabbers and the mediocre who profit from it. The corporate takeover could not have succeeded if there had been a unified front.

> Are you willing to go that far?

Yes, I grab the original sources, sometimes from Debian .orig packages. I could also use Anaconda or something, which is still corporate but less egregious.


Van Rossum worked at Google from 2005 to 2012. You think he completely avoided the structural bias to have Python solve problems Google had? Including collections.abc?

Very strange then that PEP 3119, with van Rossum as co-author, says "During the development of this PEP and of its companion, PEP 3141, we ..." and PEP 3141 is by a Google employee.

Why, it's almost as if my point was not at all about an employer interfering or making demand, but about aligned interests. As a rule, tobacco companies do not fund anti-smoker research, but also do not interfere or make demands, so those researcher can say they were not responding to pressure from their funders.

I remember the long discussion of the pros and cons of switching from Bitbucket to GitHub. The plan started in 2014, says PEP 512.

It's very clever of these Microsoft people to push for a move to GitHub some 4 years before Microsoft acquired GitHub in 2018. And to come up with a pretty good set of reasons to migrate, all documented, including a counter-proposal to use GitLab.

Almost as if it were a consensus-based decision to move, rather than a clever black-ops by Microsoft.

> I grab the original sources

Okay, then why do you care about this change or the use of PGP? Are you acting as an advocate for others, so that PyPI does not get worse for them?

In which case, why do you want them to continue to use a service where the only official way to get support is to have a Microsoft account? If you don't like PyPI, get more and more people to host their packages off PyPI.

The big gaping problem is Dependency Confusion Attacks, as in the rejected PEP 708.

Distribute your packages on a new server, using the Simple Repository API. Tell people to "pip install -i" to install your package. Take a look at the access log to see pip try to install other packages, because pip was designed for mirrors, not multiple hosting sites, each with their own potential namespace. There's no namespacing either.

From what I've read, that's pretty baked into pip, and will be hard to fix. It won't change until there are dozens of popular packages which have decided to avoid PyPI. In the meanwhile it will encourage even more people to use uv, which at least has a mechanism people can opt into.


utf8 is just one of the encodings that exist, and is kind of irrelevant at the point it contacts the llm

No, the model is trained on utf-8, so switching encodings entails a full retrain

No, the model is trained on tokens, not utf8

No, H-Net is trained on bytes, not tokens.

Congrats! That's unrelated to the grandparent comment tho.

See like something that can be handled by the multiplexer?


I'm thinking the same (e.g. extrakto). Am I missing something?


Guidelines | FAQ | Lists | API | Security | DMCA | Apply to YC | Contact

Search: