|
|
|
|
|
|
by brunorro
305 days ago
|
|
|
About doing one thing and doing it well, the best (anti) example is systemd. I think that having a DNS resolver and mounting your partitions should be two totally separated things. Fortunately nobody uses syste... Oh, wait! About the code should be understandable and the efficiency, I'm thinking of the old times java. I've seen people in systems department having to profile java stuff because they were OOMing their tomcats. Eventually we had to ask for the code, debug it and send it patched with just some '=null' because the GC wasn't doing its job in the best way. In the early 2000 the philosophy already was "Good hardware is cheaper than expert developers" :( |
|
|
People have asked the systemd folks about this, and their answer is that these are separate things because systemd-resolver is a distinct, optional binary that does one thing well. Is that enough to say they're totally separate things? Opinions obviously differ!
It is really hard to come up with rules of taste that are generally applicable yet specific enough to that people interpret them in the same way to make the same decisions. I don't really have any such rules of my own to offer. I've written coding standards for multiple projects but I definitely can't copy'n'paste one to another and have it still make sense.
> About the code should be understandable and the efficiency, I'm thinking of the old times java. I've seen people in systems department having to profile java stuff because they were OOMing their tomcats. Eventually we had to ask for the code, debug it and send it patched with just some '=null' because the GC wasn't doing its job in the best way. In the early 2000 the philosophy already was "Good hardware is cheaper than expert developers" :(
Yeah, there are plenty of examples of poorly written applications around. [1] But I don't think someone said "let's write this code really badly and just buy bigger machines". Could be wrong, just never heard anyone say that. On the other hand, I do almost universally see people prioritize the next "critical" feature over clearing the bug tracker. So if you said for example "fixing customer-reported memory leaks always takes priority over new features" that'd be making a real, controversial taste choice.
[1] ...and my guess is this was a poorly written application—keeping around reachable references too long, thus forbidding GC from cleaning them up, rather than a flawed Java GC algorithm that didn't detect the parent was also unreachable in a timely way.