4 ms·

Two weeks, Just as much as for any other new language. After two weeks new member can introduce simple features. After a month - relatively complex ones. Full

by thesz 7y ago

Two weeks, Just as much as for any other new language. After two weeks new member can introduce simple features. After a month - relatively complex ones. Full development speed is achieved after half a year and is roughly equal to all languages because all this time people learn problem(s) domain(s) and its (their) mapping to the code. For example, if I ask you to optimize Kaldi's HCLG-based decoder, you won't get any interesting results even after two months EVEN if you known C++ well. Just to tip you with something: you can save off about 10% of execution time just by presorting HCLG graph by BFS - it is scale-free and there are nodes which decoder will visit more often than others. To do this, you have to look at the graph, you have to know about scale-free graphs and their properties. Which are way outside of programming language scope.

dmitriid 7y ago
> Two weeks, Just as much as for any other new language. What?!! :) I was introducing new features within a day after switching from Java to C#. New features within less than a week when switching to Erlang (having had little FP experience). etc. etc. > For example, if I ask you to optimize Kaldi's HCLG-based decoder Why would you ask me that? Is it such a common problem? How about: if I ask you to implement a streaming Google Dataflow job getting data from a Google PubSub and writing data to BigQuery? It’s a much more common request.
- thesz 7y ago
  Oh, don't let me started on differences between Java and C#. You most probably won't write correct code in C# after day long introduction. Just to stop typing String with capital letter is a more than day long training. With Erlang you have to master OTP, which is part of language and, frankly, the only reason to choose it. The supervisor tree alone worth a week to see how it will work with different failure cases. To quote "Real Programmers Don't Use PASCAL": My first task in the Real World was to read and understand a 2O0,OO0-line FORTRAN program, then speed it up by a factor of two. https://web.mit.edu/humor/Computers/real.programmers https://web.mit.edu/humor/Computers/real.programmers It is humor, but every joke contains some truth. Optimizing Kaldi's decoder is a critical task if you recognize several thousand voice queries per minute. Again, to take analogy from outside world. Interpids mostly cruising in under-Mach speeds. It is their common problem. But they are designed to intercept other aircratfs who can fly at MachN speeds and to avoid missiles which are fly at MachN speeds. The difference between under- and over-Mach speeds is... critical? Wings of under-Mach speed airplanes would flatter at Mach1+ and destroy plane itself (see Mig15 wings). Wings of interpid will work on under-Mach speeds just okay. Common problems are easy to solve if you are ready to uncommon and critical ones. Which are often lie outside of programming language realm, can be solved easily with almost any language and Haskell provides better way to solve them (EDSLs, for example).

by thesz 7y ago

dmitriid 7y ago

> Two weeks, Just as much as for any other new language. What?!! :) I was introducing new features within a day after switching from Java to C#. New features within less than a week when switching to Erlang (having had little FP experience). etc. etc. > For example, if I ask you to optimize Kaldi's HCLG-based decoder Why would you ask me that? Is it such a common problem? How about: if I ask you to implement a streaming Google Dataflow job getting data from a Google PubSub and writing data to BigQuery? It’s a much more common request.

thesz 7y ago
Oh, don't let me started on differences between Java and C#. You most probably won't write correct code in C# after day long introduction. Just to stop typing String with capital letter is a more than day long training. With Erlang you have to master OTP, which is part of language and, frankly, the only reason to choose it. The supervisor tree alone worth a week to see how it will work with different failure cases. To quote "Real Programmers Don't Use PASCAL": My first task in the Real World was to read and understand a 2O0,OO0-line FORTRAN program, then speed it up by a factor of two. https://web.mit.edu/humor/Computers/real.programmers https://web.mit.edu/humor/Computers/real.programmers It is humor, but every joke contains some truth. Optimizing Kaldi's decoder is a critical task if you recognize several thousand voice queries per minute. Again, to take analogy from outside world. Interpids mostly cruising in under-Mach speeds. It is their common problem. But they are designed to intercept other aircratfs who can fly at MachN speeds and to avoid missiles which are fly at MachN speeds. The difference between under- and over-Mach speeds is... critical? Wings of under-Mach speed airplanes would flatter at Mach1+ and destroy plane itself (see Mig15 wings). Wings of interpid will work on under-Mach speeds just okay. Common problems are easy to solve if you are ready to uncommon and critical ones. Which are often lie outside of programming language realm, can be solved easily with almost any language and Haskell provides better way to solve them (EDSLs, for example).