3 ms·

Seems like a reasonable solution to a fraught problem. Man hours (spent recompiling the kernel with the right flags in this case) are often undervalued when co

Seems like a reasonable solution to a fraught problem. Man hours (spent recompiling the kernel with the right flags in this case) are often undervalued when comparing different solutions. I have found this is especially true when doing cost analysis. It is so easy to look at something that costs $10,000 a month and think it's expensive, despite the fact that it saves twice that much in developer time (salary).

saurik 4y ago
How many people are using Fedora? Why must all of them suffer because you want to save a very small amount of your time? Why can't the developers--which include me!!--who want to profile things take the time to custom compile their boxes instead of demanding that all of the poor end users who don't even know what "profile" means have to take a reduction in performance? How many end users would have to exist to make you decide it isn't worth their pain and extra cost for you to have to spend a minutes of your time to help them? (Maybe then further answer that question under the scenario where the users in question by and large never actually install your software.)
- prmoustache 4y ago
  Fedora End-User vs Fedora Developer edition? What prevents anyone of us to maintain a fork with same SRPMS as fedora but keeping the omition of frame pointers? I guess someone will do that if there really is an observable performance advantage. There are already fedora forks such as Nobara providing some kernel patches and some specific compilation options or patches for some apps.
- spicyusername 4y ago
  I think many many many users would happily acquiesce, knowing doing so is helping to build a better product. Also 1-2% performance hit is peanuts, given the benefit. It isn't a religious battle of ideology, it's a concrete tradeoff of something small for something large.
  - saurik 4y ago
    While I'm sure many users would, I'd be shocked if, after being asked the honest scenario of "hello: we'd like to make your device permanently 1-2% less efficient--which means everything you do on it will take 1-2% longer and your battery will die 1-2% sooner--so that, just in case it ever comes up, it will be easier for us (as it isn't like we need this, but it will save us some time and effort) we can use either remote telemetry from your device or work with you directly to make our product a bit faster" anywhere near most people agreed.
- saagarjha 4y ago
  Because that's not the hand you're dealt with. Performance problems don't just crop up on my machine with custom build flags, but on everyone's machines (yes, including end user machines!) If I get a bug report for something hanging or being slow, I can guide people through collecting a profile and sharing it with me. I can't get them to recompile their kernel and all of userspace.

saurik 4y ago

How many people are using Fedora? Why must all of them suffer because you want to save a very small amount of your time? Why can't the developers--which include me!!--who want to profile things take the time to custom compile their boxes instead of demanding that all of the poor end users who don't even know what "profile" means have to take a reduction in performance? How many end users would have to exist to make you decide it isn't worth their pain and extra cost for you to have to spend a minutes of your time to help them? (Maybe then further answer that question under the scenario where the users in question by and large never actually install your software.)

prmoustache 4y ago
Fedora End-User vs Fedora Developer edition? What prevents anyone of us to maintain a fork with same SRPMS as fedora but keeping the omition of frame pointers? I guess someone will do that if there really is an observable performance advantage. There are already fedora forks such as Nobara providing some kernel patches and some specific compilation options or patches for some apps.
spicyusername 4y ago
I think many many many users would happily acquiesce, knowing doing so is helping to build a better product. Also 1-2% performance hit is peanuts, given the benefit. It isn't a religious battle of ideology, it's a concrete tradeoff of something small for something large.
- saurik 4y ago
  While I'm sure many users would, I'd be shocked if, after being asked the honest scenario of "hello: we'd like to make your device permanently 1-2% less efficient--which means everything you do on it will take 1-2% longer and your battery will die 1-2% sooner--so that, just in case it ever comes up, it will be easier for us (as it isn't like we need this, but it will save us some time and effort) we can use either remote telemetry from your device or work with you directly to make our product a bit faster" anywhere near most people agreed.
saagarjha 4y ago
Because that's not the hand you're dealt with. Performance problems don't just crop up on my machine with custom build flags, but on everyone's machines (yes, including end user machines!) If I get a bug report for something hanging or being slow, I can guide people through collecting a profile and sharing it with me. I can't get them to recompile their kernel and all of userspace.