Join devRant
Do all the things like
++ or -- rants, post your own rants, comment on others' rants and build your customized dev avatar
Sign Up
Pipeless API
From the creators of devRant, Pipeless lets you power real-time personalized recommendations and activity feeds using a simple API
Learn More
-
Coworker: I give up! Please help me!
Me: What's up?
C: Take a look at this. I have this function here that gets the tab index and I'm passing it to the Tabs component over there. I'm logging the index and as you can see it's 3, but the Tabs component isn't working. However if I replace the function call with a 3 it works!
Coworker 2: While you were explaining all that, shellbug already thought about at least 3 reasons why that isn't working.
Me: **sighs** Of what type is the value that function is returning?
C: **stares at me for a few seconds** It's a number.
Me: Are you sure?
C: Well, it's returning 3.
Me: Please do a typeof.
It was string.8 -
Allright, I'm pissed.
Warning: more than 4k characters written by a non native english speaker ahead.
Legend:
Storytelling
> Short summary of the current situation
> "Something being said"
> (Something being thought)
* Actions *
-- Background --
In an attempt to reorganize my desktop I accidentally deleted a folder I called "development". In there I stored links to all my IDEs (Not sure how you call these in english), but also some workspaces like unity (Not much stuff there, processing (just some hobby stuff) AND Eclipse (FUCKING EVERYTHING RELATED TO SCHOOL WEB DEVELOPMENT). Now 3 days have passed and I realized this important folder was missing. Cleared that windows trash the instant I deleted the trash on my desktop.
> Shit, Regret
Install a file restore programm. Do every possible search. Nothing found.
> Big shit
Deadline was in like 3 days. Week was fucking rough so:
> "Screw this, the teacher nevet corrects the assignments and also fuck JSP"
Fast forward 2 months to last week. Teacher starts checking assignments.
> Fuck
* Sees pattern: Only students with missing or bad marks are checked. *
* Feels save *
Teacher approaching me while working on current projects.
* Doesn't feel save anymore *
> "Well, I'ld like to see your THAT programm"
> Well fuck
* Tells the truth *
> "Well that's unfortunate, but I must write a mark. Do you really have nothing to show?"
* Remember that I worked on the school pcs when I started *
> (Better than nothing. Gotta try it)
* Teacher checks programm, not pleased *
> (Fuck me, but at least it's over...)
> Nope
* Teacher calls me over *
> "With the mark I had to write today you can't reach that good mark even with a good examination, what are we gonna do about this?"
> "Well, there were other assignments that were never checked. Could we replace that mark with one of those?"
* Teacher agrees *
> (Srly bless this guy for that support)
My best choice was an Android app we had to develop during December in pairs. I did the front end (90% of the whole work) and my partner the backend (10 %). I also did 30 % of these 10 %, because I had to review the shit he wasn't able to debug himself.
> brainlogic.exe provided by windows vista
This distribution was partly my fault since I overestimated the work needed for the backend, but also the fault of that fucker. I mean, he didn't tell me the professor already provided 90 % of the backend...
Rest of the week was really busy (always 1 or 2 things to study for each day, workout and family stuff).
Yesterday (It's past 12 already) I arrived at ~9 pm in the dorm I could finally start reviewing my code.
Internet gets shut down at 10 pm.
Gotta hurry.
* Opens project *
* Sees half a year old code *
* Fights urge to puke *
> (Alright I gotta do this. For the mark!)
* waits for gradle to index files *
* Remembers the fact that I haven't opened Android Studio in the last 2 months *
For those who don't develop with android studio: This is an equivalent to ~10k windows updates waiting to be installed
> (Well, gotta work with this kinda old version)
"gradle sync failed"
> ( Ok, just restart it. You're fine )
* Android Studio doesn't react anymore and/or renders *
* Waits 5 min *
* Restarts laptop *
* Android Studio is reacting again*
"gradle is synching"
9:45 pm: gradle is done and I can finally compile my app
> FML
* Sees App launched on phone *
* Almost pukes again *
> (This was the assigment for the UX chapter, so design doesn't matter)
UX is decent. Proceeds with testing stuff. Save paths work, but some bugs can be caused by going of it
* fixes as much as possible *
* Takes quick look at backend *
Date date = new Date (GregorianCalender.getInstance().getTimeInMillis());
C'mon, I asked you to be the backend. You got 90% of the methods already written by the teacher and had 2 months to write the interfaces to my Front end AND you come up with shits like that.
Note: this example is a minor example of brainlogic.exe
I did what I could to make improve my situation. Hopefully he doesn't discover the bugs. And If it's a backend bug then I could't care less, since that was not my job!
Wish me luck for today!9 -
Did a bunch more cowboy coding today as I call it (coding in vi on production). Gather 'round kiddies, uncle Logan's got a story fer ya…
First things first, disclaimer: I'm no sysadmin. I respect sysadmins and the work they do, but I'm the first to admit my strengths definitely lie more in writing programs rather than running servers.
Anyhow, I recently inherited someone else's codebase (the story of my profession career, but I digress) and let me tell you this thing has amateur hour written all over it. It's written in PHP and JavaScript by a self-taught programmer who apparently discovered procedural programming and decided there was nothing left to learn and stopped there (no disrespect to self-taught programmers).
I could rant for days about the various problems this codebase has, but today I have a very specific story to tell. A story about errors and logs.
And it all started when I noticed the disk space on our server was gradually decreasing.
So today I logged onto our API server (Ubuntu running Apache/PHP) and did a df -h to check the disk space, and was surprised to see that it had noticeably decreased since the last time I'd checked when everything was running smoothly. But seeing as this server does not store any persistent customer data (we have a separate db server) and purely hosts the stateless API, it should NOT be consuming disk space over time at all.
The only thing I could think of was the logs, but the logs were very quiet, just the odd benign message that was fully expected. Just to be sure I did an ls -Sh to check the size of the logs, and while some of them were a little big, nothing over a few megs. Nothing to account for gigabytes of disk space gradually disappearing.
What could it be? I wondered.
cd ../..
du . | sort --sort=numeric
What's this? 2671132 K in some log folder buried in the api source code? I cd into it and it turns out there are separate PHP log files in there, split up by customer, so that each customer of ours (we have 120) has their own respective error log! (Why??)
Armed with this newfound piece of (still rather unbelievable) evidence I perform a mad scramble to search the codebase for where this extra logging is happening and sure enough I find a custom PHP error handler that is capturing (most) errors and redirecting them to these individualized log files.
Conveniently enough, not ALL errors were being absorbed though, so I still knew the main error_log was working (and any time I explicitly error_logged it would go there, so I was none the wiser that this other error-catching was even happening).
Needless to say I removed the code as quickly as I found it, tail -f'd the error_log and to my dismay it was being absolutely flooded with syntax errors, runtime PHP exceptions, warnings galore, and all sorts of other things.
My jaw almost hit the floor. I've been with this company for 6 months and had no idea these errors were even happening!
The sad thing was how easy to fix all the errors ended up being. Most of them were "undefined index" errors that could have been completely avoided with a simple isset() check, but instead ended up throwing an exception, nullifying any code that came after it.
Anyway kids, the moral of the story is don't split up your log files. It makes absolutely no sense and can end up obscuring easily fixable bugs for half a year or more!
Happy coding.6 -
Two big moments today:
1. Holy hell, how did I ever get on without a proper debugger? Was debugging some old code by eye (following along and keeping track mentally, of what the variables should be and what each step did). That didn't work because the code isn't intuitive. Tried the print() method, old reliable as it were. Kinda worked but didn't give me enough fine-grain control.
Bit the bullet and installed Wing IDE for python. And bam, it hit me. How did I ever live without step-through, and breakpoints before now?
2. Remember that non-sieve prime generator I wrote a while back? (well maybe some of you do). The one that generated quasi lucas carmichael (QLC) numbers? Well thats what I managed to debug. I figured out why it wasn't working. Last time I released it, I included two core methods, genprimes() and nextPrime(). The first generates a list of primes accurately, up to some n, and only needs a small handful of QLC numbers filtered out after the fact (because the set of primes generated and the set of QLC numbers overlap. Well I think they call it an embedding, as in QLC is included in the series generated by genprimes, but not the converse, but I digress).
nextPrime() was supposed to take any arbitrary n above zero, and accurately return the nearest prime number above the argument. But for some reason when it started, it would return 2,3,5,6...but genprimes() would work fine for some reason.
So genprimes loops over an index, i, and tests it for primality. It begins by entering the loop, and doing "result = gffi(i)".
This calls into something a function that runs four tests on the argument passed to it. I won't go into detail here about what those are because I don't even remember how I came up with them (I'll make a separate post when the code is fully fixed).
If the number fails any of these tests then gffi would just return the value of i that was passed to it, unaltered. Otherwise, if it did pass all of them, it would return i+1.
And once back in genPrimes() we would check if the variable 'result' was greater than the loop index. And if it was, then it was either prime (comparatively plentiful) or a QLC number (comparatively rare)--these two types and no others.
nextPrime() was only taking n, and didn't have this index to compare to, so the prior steps in genprimes were acting as a filter that nextPrime() didn't have, while internally gffi() was returning not only primes, and QLCs, but also plenty of composite numbers.
Now *why* that last step in genPrimes() was filtering out all the composites, idk.
But now that I understand whats going on I can fix it and hypothetically it should be possible to enter a positive n of any size, and without additional primality checks (such as is done with sieves, where you have to check off multiples of n), get the nearest prime numbers. Of course I'm not familiar enough with prime number generation to know if thats an achievement or worthwhile mentioning, so if anyone *is* familiar, and how something like that holds up compared to other linear generators (O(n)?), I'd be interested to hear about it.
I also am working on filtering out the intersection of the sets (QLC numbers), which I'm pretty sure I figured out how to incorporate into the prime generator itself.
I also think it may be possible to generator primes even faster, using the carmichael numbers or related set--or even derive a function that maps one set of upper-and-lower bounds around a semiprime, and map those same bounds to carmichael numbers that act as the upper and lower bound numbers on the factors of a semiprime.
Meanwhile I'm also looking into testing the prime generator on a larger set of numbers (to make sure it doesn't fail at large values of n) and so I'm looking for more computing power if anyone has it on hand, or is willing to test it at sufficiently large bit lengths (512, 1024, etc).
Lastly, the earlier work I posted (linked below), I realized could be applied with ECM to greatly reduce the smallest factor of a large number.
If ECM, being one of the best methods available, only handles 50-60 digit numbers, & your factors are 70+ digits, then being able to transform your semiprime product into another product tree thats non-semiprime, with factors that ARE in range of ECM, and which *does* contain either of the original factors, means products that *were not* formally factorable by ECM, *could* be now.
That wouldn't have been possible though withput enormous help from many others such as hitko who took the time to explain the solution was a form of modular exponentiation, Fast-Nop who contributed on other threads, Voxera who did as well, and support from Scor in particular, and many others.
Thank you all. And more to come.
Links mentioned (because DR wouldn't accept them as they were):
https://pastebin.com/MWechZj912 -
I was just thinking about disabling something, already forgot what it was that I was gonna disable though.. doesn't matter. And I realized that if I wanted to play my "disabled card", I could totally get Americans to ban that word entirely.
Cancel culture you say? Those cancel buttons are offensive to me! Get them out of my face reeee!
Command line? You're telling this thing what to do?! sudo make sandwich, so sexist!!!
Police reforms are so overrated. Let's ban words like master/slave or blacklist/whitelist or blind playthrough instead. And put our knees on another black person, shoot another in their sleep, and let said police officers get away with it. Yee haw!
And storm the Capitol apparently. It's been a while now but Europe looked across the pond in complete and utter surprise and disbelief. You call yourselves a free nation America?
Oh yeah, and ban words globally, in globally used software. I must've forgotten.. yeah, the world is nothing but America, oil fields, parking space and third world shitholes. Good thinking there.
With enough effort you can make anything offensive. And it goes to show that offense is not given but taken.
Fun fact btw: the United States is ranked 121 in the Global Peace Index (http://statisticstimes.com/ranking/...) - and that doesn't even include the Capitol's insurrection yet. Belgium is ranked 17. Tell me more about how I'm racist Americans. Tell me about it when your president literally called Belgium a hellhole over the amount of immigrants he saw in Brussels.8 -
I love python, but a short way to enforce interfaces on arguments would be great. I mean, this is the third function where type checking is longer than the actual body.1
-
Heres some research into a new LLM architecture I recently built and have had actual success with.
The idea is simple, you do the standard thing of generating random vectors for your dictionary of tokens, we'll call these numbers your 'weights'. Then, for whatever sentence you want to use as input, you generate a context embedding by looking up those tokens, and putting them into a list.
Next, you do the same for the output you want to map to, lets call it the decoder embedding.
You then loop, and generate a 'noise embedding', for each vector or individual token in the context embedding, you then subtract that token's noise value from that token's embedding value or specific weight.
You find the weight index in the weight dictionary (one entry per word or token in your token dictionary) thats closest to this embedding. You use a version of cuckoo hashing where similar values are stored near each other, and the canonical weight values are actually the key of each key:value pair in your token dictionary. When doing this you align all random numbered keys in the dictionary (a uniform sample from 0 to 1), and look at hamming distance between the context embedding+noise embedding (called the encoder embedding) versus the canonical keys, with each digit from left to right being penalized by some factor f (because numbers further left are larger magnitudes), and then penalize or reward based on the numeric closeness of any given individual digit of the encoder embedding at the same index of any given weight i.
You then substitute the canonical weight in place of this encoder embedding, look up that weights index in my earliest version, and then use that index to lookup the word|token in the token dictionary and compare it to the word at the current index of the training output to match against.
Of course by switching to the hash version the lookup is significantly faster, but I digress.
That introduces a problem.
If each input token matches one output token how do we get variable length outputs, how do we do n-to-m mappings of input and output?
One of the things I explored was using pseudo-markovian processes, where theres one node, A, with two links to itself, B, and C.
B is a transition matrix, and A holds its own state. At any given timestep, A may use either the default transition matrix (training data encoder embeddings) with B, or it may generate new ones, using C and a context window of A's prior states.
C can be used to modify A, or it can be used to as a noise embedding to modify B.
A can take on the state of both A and C or A and B. In fact we do both, and measure which is closest to the correct output during training.
What this *doesn't* do is give us variable length encodings or decodings.
So I thought a while and said, if we're using noise embeddings, why can't we use multiple?
And if we're doing multiple, what if we used a middle layer, lets call it the 'key', and took its mean
over *many* training examples, and used it to map from the variance of an input (query) to the variance and mean of
a training or inference output (value).
But how does that tell us when to stop or continue generating tokens for the output?
Posted on pastebin if you want to read the whole thing (DR wouldn't post for some reason).
In any case I wasn't sure if I was dreaming or if I was off in left field, so I went and built the damn thing, the autoencoder part, wasn't even sure I could, but I did, and it just works. I'm still scratching my head.
https://pastebin.com/xAHRhmfH33
Top Tags
Weekly Rant
View