Re: [lmi] Include what you use

[Greg Chicares]

On 10/27/21 12:06 PM, Vadim Zeitlin wrote:
> On Tue, 26 Oct 2021 23:41:12 +0000 Greg Chicares <gchicares@sbcglobal.net> 
> wrote:
> 
> GC> [...] In some cases it duplicates what
> GC> the source already says:
> GC> 
> GC> iwyu: #include <climits>            // for UCHAR_MAX
> GC> lmi:  #include <climits>                      // UCHAR_MAX
> GC> iwyu: #include <cstdio>             // for EOF
> GC> lmi:  #include <cstdio>                       // EOF
> GC> 
> GC> but it's very nice to have an automatically-generated touchstone
> GC> to verify that.
> 
>  FWIW I thought about converting its output to lmi-compatible form by just
> passing it through sed/awk/rakudo/deep-learning-AI to get rid of "for" and
> start the comment at 41st column -- this would be simple to do, but mostly
> (only?) useful if we wanted to use it automatically, e.g. to compare the
> lines output by the tool with the actual lines in the sources.

I don't think we'd want to use it in quite that way. For example:

  #include <string>             // for string

That's obvious. I do like to be able to press '*' (shift-8) while
the cursor is on "string", in order to judge whether <string>
really should be included; but I can do that without the comment.

Similarly, I'd omit this comment and press 'g*' on "limits" here:

  #include <limits>             // for numeric_limits

Then, if 'g*' finds nothing, that tells me I should consider
not including the header.

Here, I'd have to think about which "operator|' is meant:
  #include <ios>                // for ios, operator|, ios_base, ostringstream
and I'd still want to include <sstream> for std::ostringstream.
The tool's greatest usefulness here is to make sure that only
appropriate headers are included; I don't spend much time
thinking about what's in class std::ios std::ios vs. std::ios_base.

> GC> > miscellany.cpp should add these lines:
> GC> > struct tm;
> GC> 
> GC> I wonder why it would complain here:
> GC>     std::tm const*const t1 = std::gmtime(&t0);
> 
>  Sorry if I'm misunderstanding your question,

I thought it was suggesting that, in order to use std::tm,
merely including <ctime> wouldn't suffice, and that it
would also be necessary to declare struct tm explicitly.

> but to me it seems clear
> enough: the tool wants to forward declare std::tm because it must be
> declared here and forward declaration is enough, i.e. it's unnecessary to
> include the full <time.h> here.

But would a simple
  struct tm;
declaration suffice? And shouldn't that be
  namespace std {struct tm;}
? But wouldn't that full declaration be illegal? And wouldn't
we still need <ctime> for the declaration of gmtime(), or does
this tool want me to forward-declare that, too?

In less time than it takes to pose those questions, let alone
answer them, I can simply type
  #include <ctime>
and know I'm in full compliance.