On Mon, Dec 26, 2016 at 2:58 AM, Vladimir Batov
On 2016-12-26 10:41, Andrey Semashev wrote:
... Acknowledging the need for a particular functionality is not the same as accepting a particular implementation.
Put bluntly users do not care about implementation. They care about functionality provided.
They don't until they hit the issue. I think there is a difference between generic utility library like "unordered" and domain specific library. When we talk about domain specific library - users expect that its "under-the-hood" implementation would meet the highest standards - so they may think of API. Especially when it goes to doing something cross platform. Doing the review I felt that the Windows backend is well written for MSVC but the rest was lacking and was implemented in as stop-gap solution to provide the functionality for Linux. Especially that it wasn't mentioned in any place in docs how it is actually done. (@Antony Polukhin I'm sure you worked hard on it and I apologize if it offends you somehow) But as domain specific library I think there is some way to go. And that is why the original vote was conditional yes. The problem is that the backend needs rework, now with additional issues in frontend I decided to change the vote - because in its current shape. But I do want it to be back for the review after updates and I clearly stayed so.
As for the feedback, then I do not see voting something useful "no" to be much of a feedback.
Indeed - that is why this library received lots of feedback - so the author can do fixes in design and come back.
"difficult", "painful", "backward compatibility"... Come on. I am sure you've been in the industry for a while, have "survived" many upgrades/updates, etc. We all know it's not an issue. We update, adjust, move on.
I must say the poor - or more correctly say non-existant backward compatibility of the boost libraries limits its usefulness significantly in real world industry. I've seen companies that: 1. Stuck with old versions of boost because they can't upgrade without breaking the entire code base 2. Limit themselves to very small parts of boost Artyom Beilis