I'm experiencing what I think might be a bug with mysql. It's certainly weird.

I have a query something like

In my test, about half the time, one of the ids is returned twice. Only one, always the same one. The id is contained in both subqueries (but so are other ids)

If I remove BOTH left joins, it will return only one result. Removing either one makes no change. Neither of the left joined tables have *anything* in them, in this case.

If I replace the subqueries with literal lists of ids, there's no problem.

Each time I re-run the test (which starts with a fresh database), it either works or it doesn't. If it works, all subsequent queries will work, if it doesn't, they won't. So it's something about the structure of the data, I think, not (just) the query.

Well, ok, I just ran it like 10 times and it failed every time, so it doesn't work 50% of the time. I guess I got lucky before.

Over time this should become less and less frequent to the point it almost never happens, or else you're doing something wrong.

I have lust in my heart for the space gray 27" super iMac Pro starting at only $4999.

ya I mean, its a joke "well, it works on my local". thats not good enough.

wow that sounds awful.

verifying identity, just to the point of uniqueness and without being invasive, needs a better solution than those available to us. couple imperfect ideas i've used in the past that may repel some of your trolls (though sadly some real users too, but if it gets bad enough may be worth it):

verify phone number (send sms confirm link via twilio or similar)
force low-permission facebook/google account link to create account. (both of them make it at least somewhat annoying to create fake accounts, so you'll effectively be offloading work to them)

Cheaper than what I just built.

There is a reason that chat isn't common and it isn't because someone can't over engineer a solution to a problem that shouldn't exist.

i'm not clear, are you saying public chatrooms are just a bad idea, period?

Yes.

The amount of engineering to going into a basic obs chat (which generally serve no purpose other than to waste your time and provide an avenue for spammers/griefers to do their thing) is absurd. Unique ID, filtering, karma system, throttling, etc. And this is just to prevent basic abuse much less drive desired outcomes. (The magical "engagement" that people think this fosters.)

Is this the first apple shipping with 32GB ?

Not sure, haven't been tracking hardware for awhile. Thought they said the mac book pros could go that high.

Can you please extrapolate on why that is? Thanks.

Well Grues games is definitely going to attract the 4chan alt righties because there's a political element to it.

Over time you should be getting your personal machine in sync with the dev environment. If you let it happen twice you probably didn't try hard enough after the first time, or your environment is changing without notice which is also bad.

Suppose that I have written two solutions to an exercism.io problem that both pass all the required tests. now i want to see which solution is faster.

theoretically, how can i go about testing 'em to see which one is faster?

when i run them in my IDE, i get a wide range of time responses just running the same code. so i might run it five times and get .34s, .45s .7s .29s .98s.

A typical solution is to run several times and take the minimum. The min should represent the best your algorithm can do - higher values are most likely from fluctuations outside your control - other processing getting swapped in, disk activity, etc. It also can help to increase the amount of times your algorithm runs per execution, because there's some "startup cost" to starting any program you write, especially if it's a scripted language (python, ruby, perl, whatever) that can be very variable

Definitely agree about running many times (and with varied input), but I don't think I've seen the minimum being used to compare algorithms before. Depends on the use case, but typically I see/use median, 95th percentile, 99th percentile, or possibly mean.

redux is pretty awesome. really like being able to save state.

Great post today about stock options

https://danluu.com/startup-options/

whatever language you're using will have benchmarking facilities, or else third party libraries that provide them. those facilities will run the code blocks you're comparing many times and average them. you could write your own version of that, ofc.

Just wondering theoretically if it's possible to count the assembly instructions or something to get a consistent and completely objective result?

It's probably one of those things that IS theoretically possible - but is practically impossible with our current knowledge / technology.

It's not just assembly instructions - it's also how the instructions are able to be processed/optimized by the hardware. The actual data being processed would influence the result as well. Ordered or mostly-ordered data might perform very differently than random data. Things like resource usage would matter as well. A resource intensive algorithm might perform better theoretically based on pure code - but perform poorly in practice because of the machines the its running on or based on other programs that are running.

I think there are just too many factors for us to account for to get a consistent and completely objective result in most cases.

Sounds like a job for machine learning! Let the machines tell us how fast they are, what could go wrong?

I do wonder how much this matters though. The numbers he's showing seems within a typical deviation and I'm guessing this is all small inputs. Taking minimum(A) -vs- max(B) would be bad, of course.

Discounting manually checking Big O, etc.

I first heard of the idea of using the min from the python timeit module docs, and it makes sense to me - provided, I guess, that each time you run it, it's doing "the same thing", like, sorting more or less the same list, or whatever. If you're randomly choosing the length of the list or the order of the list, then no, something like average will be more useful.

But if you're doing "the same" thing over and over, then the minimum should represent the best the algorithm can do, without pollution from other things in the system you don't have much control over.