Re: [lmi] expand_html() msw64: 'SIZE_MAX <= index' always false

[Greg Chicares]

On 2019-03-24 20:41, Vadim Zeitlin wrote:
> On Sun, 24 Mar 2019 19:18:17 +0000 Greg Chicares <address@hidden> wrote:

[...string manipulation with std::strtoul()...]

> GC> I figured that you had chosen SIZE_MAX because you wanted type 
> std::size_t,
> GC> much as we compare std::string::find() results to std::string::npos.
> 
>  But find() is documented to return npos, while std::strtoul() is
> documented to return ULONG_MAX in case of overflow. And I [know that I]
> know it. So this can't be the explanation -- but the commit 416ab02f that
> you have so obligingly found for me does provide one:
> 
>  In this commit, the check preceded a static_cast of index to size_t, so
> I must have just decided to make it absolutely manifest that the cast was
> valid and didn't realize that this could result in the problem when
> ULONG_MAX < SIZE_MAX.

Without looking at the history, I'll assume that I saw std::size_t
and naturally vaporized it without a second thought, presumably
without replacing SIZE_MAX. If so, I didn't miss any other occurrence:

/opt/lmi/src/lmi[0]$grep --quiet SIZE_MAX *.?pp || echo "no occurrences found"
no occurrences found

so the mystery has been resolved and everything's now okay.

But why use that C function anyway, instead of std::stringstream?
Granted, std::string and std::stringstream aren't as good as they
could be if they were redesigned today, but OTOH when I examine:
  /opt/lmi/src/lmi[0]$grep strto *.?pp |less -S
I see your comment here (which I heartily agree with):
  rate_table.cpp:// Helper function wrapping std::strtoull() and hiding its 
peculiarities:
and great pains are taken in 'numeric_io_*.?pp' to hide those same
peculiarities; this
  pdf_command_wx.cpp:  auto const index = std::strtoul(s.c_str() + open_pos + 
1, &stop, 10);
is the only line that uses any of these functions directly.