What was a project or spec that got put on your desk that could not possibly be done? How did you explain the dilemma to the "requester"? More importantly, did they understand after you explained the fundamental issue?
Not assigned to me per se but just what my thoughts were on...
"Could we build [something that mimics Excel functionality] as an online product, in a few months time?"
Online spreadsheet functionality. The thing that the two greatest software companies in human history (Microsoft and Google) are working on. And you want to create a basic engine plus functionality that doesn't currently exist in either. In only a few months time. Using the talent we have in-house.
I was asked to figure out a way to determine when and if a thread would stop. This was so we could find and prevent all these multi-threaded bugs. Guess they'd never heard of the halting problem. (then again, they also didn't believe in locks, mutex's, or using any sort of synchronization methods.)
I was asked to implement search, in 2 days for our enterprise app... not too bad. I can do that. I specifically noted:
Just so you know, I'm not google, so the fancy ranking and auto-suggestions and all of that... not likely gonna happen, especially not in 2 days.
3 days later: "But google does..."
Once I was asked by an old, nearly senile man to make a program to win a lottery.
I was like...
"Oh... (hahhahahahahahahahaha) sorry, it's just allergies, but go on..."
He showed me his "theories" he came up with after analyzing 20 years worth of records and that if I just made a program that followed his rules, he would pay me.
I made the program, gave it to him, showed him it worked according to his rules, got paid, wished him good luck and told him if he ever bet too big and lost too much money, it's not my fault in any way.
I and another couple of programmers were once asked why we couldn't find an algorithm to determine where holes were in 3-D geometries composed of triangles (STL files, for those who know them, and not exactly conforming to the proper specification of no-holes-allowed). My boss didn't seem to understand how hard it could be, because after all, he could just look at the rendered model and point to the holes in it.
Once I was tasked with creating a 1:1 translation of Access VBA code to PL/SQL. Things went relatively well, until I found out that deep within several nested IFs and LOOPs of a procedure, it would open a yes-no-dialog and ask for another decision by the user. The best I could offer, and eventually implemented, was to add another parameter to that procedure; if it ran into that branch, it would rollback, return a special status that instructs the GUI program to ask the user, and then restart the procedure with an extra parameter.
(not me, but a co-worker) Long ago, I worked building Windows software that could control and monitor industrial machinery. One option for connecting to the machinery was to dial in via a modem.
A large customer called a high-level manager with a problem: They could not connect to a remote facility because they kept getting a busy signal (they were sharing the hardware line with a voice line or something)
When my coworker found that the problem was not a software fault, but that the phone line really was busy, the manager responded "That's ridiculous. Can't you code around that?"
I once was asked to fix regex that was being used to parse C# function declarations because the QA team had found a way to break it. This unfortunately is not a possible task as function declarations cannot be described as a regular language in C# (attributes, generic types, attribute arguments, etc ...).
I resolved the problem by simply explaining that it was an impossible task and detailing a few examples with nested generic types. Management didn't believe me at first (was 2 weeks into the job) but I was able to explain the problem to a couple of senior developers and they in turn convinced management. Eventually they decided to have a best effort regex and add a real parser in later.
Automatically generate a list of all valid search combinations that return results from the database. While not logically impossible from the standpoint of coding, given that the size of the database was always growing, it was highly unpractical and the sever would eventually slow to a crawl while trying to generate combinations.