HNHacker News
TopNewBestAskShowJobs

charleslmunger

1,327 karma · joined March 3, 2013

submissionscomments
charleslmunger··on Android vendors, don’t kill my app
Android doesn't offer any signal of normal termination. Your process gets killed when the OS needs memory for some other app.
charleslmunger··on Start with a Website, Not a Mobile App
Twitter's mobile web site is the most unreliable site I encounter with any frequency. Fully 80% of the links I get to Twitter fail to load, telling me to try again, which only works when I fully refresh the page.

I don't know how it compares to their native app.

charleslmunger··on Audiophiles in Japan Are Installing Their Own Power Poles
Yes, but when when there's signal degradation in an HDMI cable it shows obviously, not as a subtle video quality change. If you can see an HDMI signal with white blinking dots everywhere, you can be confident that the cable is not altering the signal in any really detectable way.
charleslmunger··on A $1.2B law to boost US quantum tech
NIST thinks that 2048 bit RSA will be secure until 2030. Presumably their analysis does not include a quantum computer, because if it did they wouldn't recommend 3072 bit RSA for security beyond 2030. Their analysis is indexed against DES (so cracking 2048 bit RSA using the best known methods would be equivalent to cracking a hypothetical 112-bit DES).

It's possible that you're right that a quantum computer won't exist in 2040 that will break 2048 bit RSA, but that doesn't mean it won't be broken anyway by then.

charleslmunger··on “Stylish” is back, and you still shouldn't use it
On Android, camera is a separate permission from reading storage. In addition, the permission is not even required if you're just getting a photo from the user - your app can launch the camera app, and get a callback when the user has taken a photo, with access only to that photo.
charleslmunger··on US asks allies to drop Huawei
Also not wrong. But the US, while not a dictatorship, and with its own share of problematic foreign policy, has been caught with its hand in the security cookie jar.

DUAL_EC_DRBG was included in BSAFE and Juniper products. So even if we assume the worst of Huawei, it's really a matter of which back door you want in your networking equipment. My hope is that these constant accusations make Huawei drive big improvements in reproducible builds, source-available software, and verifiable hardware. But I'm not holding my breath.

charleslmunger··on US asks allies to drop Huawei
To be fair, their main competitor, Cisco, has had a significant number of hardcoded password backdoors found in their products. While their reporting and patching story is better, their security track record is still awful. It's believable that Huawei as awful security and backdoors, but that is just as bad as Cisco. The real difference would be if they were actively aiding espionage, which has been talked about a lot but never backed up with evidence.

https://www.tomshardware.com/news/cisco-backdoor-hardcoded-a...

charleslmunger··on Use `make` to save time
Maven is possibly one of the worst build systems I've used. It's an imperative build system with a declarative syntax; not parallel by default; always requires a clean build in practice, unless your code is all stuffed into one big blob. For a long time, it happily downloaded executable code from the internet without even basic cryptographic verification.

Even the ergonomics of the tool are poor - it spams log output with all sorts of information other than what you actually need (like the output of a failed compilation). Getting information from a failed test is painful. If you change one module that another depends on, the latter will not be rebuilt - letting you find issues via NoSuchMethodError at runtime. Annotation processors require special care to enable, and don't always run when they should.

I would rather use make and javac than Maven. Gradle, Bazel (and its diaspora of non-google imitations) would be much better choices.

charleslmunger··on Big List of Naughty Strings
I have no knowledge of apple's decisions here, and no experience with iOS, and you're generally correct that any sort of batch processing should be able to recover from bad input, but I'll try to make an argument as to why this isn't as simple as it sounds.

1. Creating a graceful degradation path is a complicated and expensive, from a UX, product, and testing perspective - if you silently drop the message, that's much worse than crashing; if you insert some sort of tombstone "this message could not be parsed" into the UI, you have to figure out if that's something users will actually understand. These code paths are hard to exercise with normal integration or manual testing, and it's likely that over time they may stop working correctly.

2. Monitoring graceful degradation is more complicated than tracking crashes, especially for unforeseen issues like what you're describing. There's a real risk that if this periodically showed up with "failed to parse" messages, the actual issue would have remained undiscovered by Apple a lot longer.

3. Again, no familiarity with iOS, but on other mobile operating systems I've worked with there's a significant memory overhead to using multiple processes and IPCs this way. If your device is under memory pressure, the extra cost of a pipeline like this introduces new failure modes. First, your other process may be killed by the LMK, which the sender process could interpret as being a failed message. Second, you may increase the amount of memory required to receive a text message in the background - this can directly affect critical high-memory use cases like taking a live video with the camera. There may just not be enough room to do both at the same time.

4. There's significant input processing that can't be done in another process, or meaningfully isolated from that of other messages - the best example being UI rendering of the text. If there's a magic string that causes view measurement to fail, that's extremely difficult to attribute to any specific piece of input - so adding this extra process to do validation won't really help you, since the "validated" string will fail later on.

charleslmunger··on Sweden's cashless society is no longer a utopia
I suspect this allows the festival organizers to more easily take a cut from the merchants.
charleslmunger··on Avoiding race conditions in Swift
Absolutely this. See https://google.github.io/guava/releases/23.0/api/docs/src-ht...

For a discussion on some of the dangers of invoking completion listeners synchronously.

charleslmunger··on The European Union versus the Internet
From the link:

makes available to the controller all information necessary to demonstrate compliance with the obligations laid down in this Article and allow for and contribute to audits, including inspections, conducted by the controller or another auditor mandated by the controller.

Adherence of a processor to an approved code of conduct as referred to in Article 40 or an approved certification mechanism as referred to in Article 42 may be used as an element by which to demonstrate sufficient guarantees as referred to in paragraphs 1 and 4 of this Article.

You can subcontract, in the same way that a any other business has to subcontract with businesses that obey relevant laws. They didn't ignore history or the present - they added new responsibilities to subcontractors, and described requirements for those contracts.

The subcontracting provisions I think are actually very reasonable and well defined. Things like the Right to be Forgotten have other issues around free speech, but the controller -> processor relationship seems pretty well specified.

charleslmunger··on Research shows that nearly every U.S. cell provider is doing throttling
All of those are not supported by the evidence in the linked article. They observed specific fixed speeds per provider, regardless of location, time, or load. You're correct that there are emergent effects from peering, routing, etc, but that is not what the evidence suggests is the cause of the issue.

And the specific regulation of "Do not throttle or shape legal traffic based on origin or contents" doesn't require large complicated regulatory intervention. A simple complaint system "we observe that traffic to origin X exhibits dramatically different performance than traffic to origin Y" and response would work basically fine. Of course it would be better if ISPs would behave reasonably without regulation, but we can see clearly that it's not the case.

charleslmunger··on The European Union versus the Internet
I am not a lawyer, but GDPR explicitly covers the plugin pipelines - they're "processors". The requirements for processors are basically that you can only use processors that are compliant with GDPR themselves. Any well designed regulation disallows skirting liability by subcontracting out functionality. Is that really unreasonable? It describes pretty clearly how to be a compliant processor, and it's basically saying that you have to have a contract with the "controller" that requires you to fulfill the same responsibilities that the controller would have under GDPR if they were doing the work in house.

https://gdpr-info.eu/art-28-gdpr/

charleslmunger··on Research shows that nearly every U.S. cell provider is doing throttling
I'm not sure I follow your logic - it doesn't take a deep investigation to discover throttling, it takes a deep investigation to prove that poor performance is caused by throttling. It doesn't take a deep investigation to see that your cash register is short, but it might take one to prove that an employee is taking money from the drawer.

And we're not seeing network capacity greatly increased through sensible traffic shaping - that's an assertion made without evidence. If that were the case, we'd expect different video providers to be shaped the same way, and we'd expect the throttling to change based on time of day and peak traffic, which the researchers did not observe.

charleslmunger··on EU approves internet copyright law, including ‘link tax’ and ‘upload filter’
IANAL, but the legal answer for why free speech and copyright aren't mutually exclusive is that copyright is an explicitly listed power of congress in the constitution:

The Congress shall have Power To... ...To promote the Progress of Science and useful Arts, by securing for limited Times to Authors and Inventors the exclusive Right to their respective Writings and Discoveries;

and

Congress shall make no law... ...abridging the freedom of speech, or of the press;

So it's implied (by the constitution not being tautological) that there's a balance between the right of the people (free speech) and an power of congress (to implement copyright). Note also that that defamation/libel are not protected by the first amendment, but the theshold for proving it is high - information has to be known to be false to the speaker, and inflict an actual injury. For public figures and politicians, the speaker has to have "actual malice". These are not often proven. For example, "sell more newspapers" would not rise to the level of "actual malice".

charleslmunger··on Securing MongoDB Using Let's Encrypt Certificate
If you tell your clients to allow self signed certs, your connection is vulnerable to a MITM, which is usually what you're trying to avoid by using TLS...
charleslmunger··on NSA metadata program “consistent” with Fourth Amendment, Kavanaugh once argued
See my other comment:

https://news.ycombinator.com/item?id=17949047

charleslmunger··on NSA metadata program “consistent” with Fourth Amendment, Kavanaugh once argued
This article [1], from a source that is generally quite positive on Kavanaugh, provides much more evidence of a trend in Kavanaugh's interpretation of the 4th amendment from his previous cases. That's what originally informed my opinion.

[1] https://reason.com/archives/2018/07/18/brett-kavanaughs-four...

charleslmunger··on NSA metadata program “consistent” with Fourth Amendment, Kavanaugh once argued
Yes, but the question we care about here is if we'd get a privacy preserving decision like Carpenter if Kavanaugh was on the court. I read that section of his opinion, and it certainly seems like the answer is "no".
charleslmunger··on Paper, the least terrible password management tool (2015)
Paper is an awful medium for password management, for a host of reasons:

1. It's vulnerable to phishing. The vast majority of sites still do not support U2F or similar, and typing in a password from paper means that a human is validating the domain, not a machine.

2. It's not encrypted at rest. This means that leaving your wallet somewhere means all your passwords have been exposed. It also means in many legal cases (IANAL, US law, etc) the contents of that paper are a key to all the other encryption you might have. In particular, courts have held that the fifth amendment protects a defendant from being compelled to reveal passwords in many circumstances.

3. It's difficult to keep in sync with new registrations, logins, and password changes - given the widespread use of mobile phones, it means that you are likely to take that piece of paper with you, compounding the risks in #2

5. Paper isn't durable. Spill a drink on it and everything is basically gone.

6. Paper isn't generally structured to obscure passwords - open your browser's password manager, and you'll see for each password unless you click to reveal it. With paper, you are much more vulnerable to shoulder surfing.

Edit:

7. Any decent password manager will volunteer to generate passwords for you - if you're doing this on paper, it's much more likely that you'll try to come up with a password yourself, resulting in weaker passwords.

charleslmunger··on Apple self-driving car in accident: California DMV filing
> Braking systems are controlled by a dedicated embedded software system that's extremely well-understood and well-tested (probably even formally verified).

Are you sure? That code is closed source, and the peek behind the curtain we saw with Toyota's investigation did not inspire confidence:

http://www.safetyresearch.net/blog/articles/toyota-unintende...

charleslmunger··on Intel Publishes Microcode Patches, No Benchmarking or Comparison Allowed
How much of that boost remains after patches to Spectre and meltdown?
charleslmunger··on DoorDash Valuation Nearly Triples to $4B in Six Months
Investors are willing to pay higher prices for less equity for two reasons:

1. Normal investment of $Y at valuation X implies that you, the investor, think that the company is worth more than X. If it exits for more, you'll get more, less you'll get less. But investing $Y at valuation X with a 2x liquidation preference means that you think the company is worth merely Y. The valuation basically doesn't matter any more. If I invest $10 at a $100 valuation and the company exits at a $500 valuation, I get a $40 profit. If it exits at a $50 valuation, I've lost $5. The valuation I invest at makes a big difference for my profitability. But consider the same scenario with a 2x liquidation preference - in the happy case, I get the same profit. In the sad case, I've still doubled my money.

2. Investors in startups are not perfectly rational investors. If I own an existing stake in a company, I would fight tooth and nail to prevent any new money coming in from having a liquidation preference over me. But in truth, if the company can raise money at a $2 billion valuation normally, or a $4 billion valuation with a preference, a fund can report double the gains to the people who actually put money in it. So they stay quiet, show off the impressive paper returns of their current investment fund, and go on raising the next one. This effect is doubly pernicious when the later investors are also earlier investors - they can basically juice their own balance sheet. On paper, the existing stake is worth more because of the inflated valuation, and the new stake is a heads-I-win-tails-you-lose.

And at the lowest preference are all the employees. Founders take money off the table in the inflated new rounds, so while they technically have a lower preference they've still cashed out significantly.

charleslmunger··on SQLite – The “server-process-edition” branch
The link says that it requires turning synchroous off, which means that you won't be waiting for real I/O on transactions, since no fsync calls are emitted. Add that to a huge amount of RAM for cache, and it's very reasonable to be CPU bound... So long as you don't mind data loss or corruption on power loss or kernel panic...
charleslmunger··on SQLite – The “server-process-edition” branch
If this requires turning off pragma synchronous, doesn't that disqualify it for anything but read only tasks? Power loss etc would corrupt the database...
charleslmunger··on Google Public DNS turns 8.8.8.8 years old
Just some related stories:

https://www.eff.org/deeplinks/2014/11/verizon-x-uidh

http://www.latimes.com/business/la-fi-lazarus-20150818-colum...

There is no point in discussing what companies might do in some theoretical framework - they do not care about your privacy, they monetize NXDOMAIN, etc. This isn't about what might happen - these abuses have already happened.

charleslmunger··on Google AMP – A 70% drop in our conversion rate
Is "Not Secure" a false label? Plaintext HTTP is literally not secure against either a passive or active MITM.
charleslmunger··on Is Apple Really Your Privacy Hero?
On android, it's possible for apps to access contacts in a few ways:

1. Request permission to access all contacts at runtime - user chooses yes or no.

2. Ask the user to choose a specific contact. This pops up some system UI, and then relays back the information the app requested about the contact the user chose:

    public void pickContact(View v) {
        Intent contactPickerIntent = new Intent(Intent.ACTION_PICK,
                ContactsContract.CommonDataKinds.Phone.CONTENT_URI);
        startActivityForResult(contactPickerIntent, RESULT_PICK_CONTACT);
    }
    @Override
    protected void onActivityResult(int requestCode, int resultCode, Intent data) {
        if (resultCode == RESULT_OK) {
            switch (requestCode) {
                case RESULT_PICK_CONTACT:
                    soSomethingWithContact(data);
                    return;
            }
        } else {
            Log.e("MainActivity", "Failed to pick contact");
        }
    }
And for most cases, it's also fine to just use the "share" intent to send a link or similar via whatever app and whatever contact the user chooses, without any extra info leaking to the providing app. You can actually see this in action with the uber app - if you choose to share your ETA with someone, it first asks for permission to access all your contacts. If you deny it (which I do) then it follows up by having you send a link via the regular share intent.
charleslmunger··on Voting system to be used in West Virginia elections is vulnerable
There's still the same dispute about miscalibrated touchscreens on electronic voting machines - the difference is that paper provides an audit trail, such that it can actually be disputed.

https://www.npr.org/2016/10/26/499450796/some-machines-are-f...

← PreviousPage 9 of 16Next →