On Tue, Feb 27, 2024 at 4:55 AM Dan Eble <nine.fierce.ball...@gmail.com> wrote:
> On 2024-02-27 07:14, Luca Fascione wrote: > > inline constexpr vsize VPOS (vsize(-1)); > > OK, but don't repeat the type. > > inline constexpr auto VPOS = vsize (-1); > Well. To that, yes and no. I read it better in the original way, and repeating the type at a distance of less than 10 characters is ok IMO (in that you obviously see it if/when you're changing the code, so parsing the 'auto' is substantially more cognitive overhead than reasoning about the type being there twice). Besides, depending on how good code intelligence your editor has, I want to see the type in the popup ballon describing VPOS at the usage size, and finding 'auto' there will just make me frustrated. In my school of thought auto should only be used where it's necessary (like inside templates) or in places where the type declaration is so convoluted that reading it is not useful (like std::map<complex type, another complex type>::const_reverse_iterator::value_type kinda things). > Personally, I don't like flower's aliases related to vsize, > ssize, and Real -- I'd rather just use standard names -- > but I remember that there was some resistance to changing > them the last time I suggested it. > Redeclaring size_t is a thing of a very distant past, I've worked with APIs that had declarations for int and float (and even void in one, yuck). It's not a modern way of working, but if the thing is external you should use types that match the API definitions, that'll make it easier to keep your code correct when the typedefs move (like when you change platform, the whole long vs long long thing on linux/windows is a complete nightmare, for example) I know that at the moment a native build on windows is not on the cards, but it seems to me that adding friction gratuitously is just an unwise thing to do. But to me, if you have a variable that is intended to be passed as a parameter to a thing that wants a foo_t, even if you know that foo_t is a typedef for bar_t, your variable should be declared as foo_t, you'll find bugs faster that way, because you're pulling the detection of mismatched types closer to code you can act upon. L -- Luca Fascione
