2 ms·
> I like to think of Xcode's syntax checking and highlighting as "eventually consistent". This made me laugh, but I suppose I've been spoiled by the responsive
by notduncansmith 12y ago
> I like to think of Xcode's syntax checking and highlighting as "eventually consistent".
This made me laugh, but I suppose I've been spoiled by the responsiveness of text editors. Honestly I wish there were a way to turn off the error checking without losing autocomplete as well.
> switching to vi for code editing while still using Xcode for the rest doesn't seem like it would be an improvement
Agreed - I tried both Sublime and vi alongside Xcode, it's just as (if not more) cumbersome than Xcode alone. I wouldn't even really mind the sluggishness of Xcode if it offered the same text editing capabilities (multi-cursor support, etc) as Sublime. At least then the problem can be reduced to buying a beefier machine.
- MichaelCrawford 12y agoQuickLetter from Working Software ran just fine on an 8 MHz Mac Plus with 4 MB of memory. I expect it would have been able to run, perhaps not acceptably but it would have run, with just 2 MB of memory. It had a full-featured styled text engine - fonts, tabs, margins and so on. It was built on the same CoreEdit styled text engine that MacWrite was. The very first MacWrite, I'm pretty sure ran on the 1984 Superbowl Mac - that model was later given the name "Mac 128k". The Mac 128k had a 64kB ROM and single-sided 400kB floppy disk; there was a port for an external 400kB single-sided floppy however most folks didn't have one as the drives were quite costly. Fast forward to today, and I have a 64-bit Core Quad i7 CPU, a 500 GB SSD drive, 8 GB of DDR3 memory, and all the advances that academic computer science has made in the way of algorithms but even so... it is quite uncommon that I don't regard a GUI program - open source or proprietary - as unacceptably unresponsive to my mouse or keyboard. Consider Moore's Law. I don't think he actually said that hardware doubled in speed every 18 months, rather that the transistor count per chip doubled. But loosely speaking that produces the effect of hardware getting twice as fast. It is incredibly difficult to double the transistor count on an already complex chip. How hard could it possibly be - to make your executable code run half as fast? :-/