Add fmt for string formatting

[SirNate0]

Implementation of request #9. Adds the fmt library for a FormatString utility with {}-style format to eventually replace the printf style formatting with %s, %d, %f, etc used by ToString("format-str",...). Further, adds U3D_LOG____F macros that use this fmt-style format string, while the URHO3D_LOG____F macros use the printf-style through a deprecated call to Log::WriteFormat (use Log::WriteFormatted for the {}-style).

---

Before this is merged, I would like to:

• Upgrade to a more recent version of fmt (version 9 rather than the 5-year-old version 3)
• Convert the warning-generating calls to sprintf in Profiler.c to use this new formatting code. Supposedly, fmt is faster than printf, at least on release builds, so this should not be problematic in terms of performance.
• Have some of you test it, as I have only tried it on my Linux system.
• Double check that I haven't messed up the line endings on the files. (What is our desire for the code - LF or CRLF?)

[SirNate0]

The web build fails, as the tools are built with an incomplete "Mini" version of Urho, which I did not set up to include fmt. I will go ahead and add it as a dependency to them if there are no objections.

[eugeneko]

By the way, you can avoid adding new macros for logging. We rewired LOG macros to format with {} syntax by default, and LOGF with % syntax, both are using fmt of course

[SirNate0]

That is an excellent idea. I will switch to that instead.

With a template specialization or overload, it should be possible to still call a completely unformatted version of the logging code, so no extra runtime cost as well.

[SirNate0]

Implemented Eugene's suggestion. Also changed to fmt 9.1, and since those were fairly comprehensive changes, I just started from a clean slate so the history is cleaner.

To upgrade the log macros, I used the following script. It may be useful to some users, so I'm putting it here, or we could add it under script/utils.

#!/usr/bin/env python3

import re
import os
import sys
import re
from glob import glob

print(sys.argv)

bad = []

def convert(f):
    if f.name.endswith('Log.h'):
        return None
    content = f.read()
    # URHO3D_LOGF is only used in the definition, so no need to specially handle it with a non-string first argument
    for level in ['TRACE','DEBUG','INFO','WARNING','ERROR','RAW']:
        matches = list(re.finditer(f'URHO3D_LOG{level}F'+r'\(((\s*"([^"]|\\")+")+)'
                        ,content))

        calls = list(re.findall(f'URHO3D_LOG{level}F',content))
        if len(matches):
            print(f.name)
        if len(matches) != len(calls):
            print('ERROR, could not convert all of the calls in ',f.name)
            bad.append(f.name)
        for m in matches:
            print(m)
            print(m.group())
            #print(content[m.start():m.end()])
            newstr = m.group().replace('{','{{').replace('}','}}')
            for c in 'diulfcsxp':
                preserve = 'cx' # Assume remaining are handled correctly by default: (unsigned) integers/longs, floats, strings, and pointers
                if c not in preserve:
                    newstr = re.sub(r'([^%])%'+c,r'\1{}',newstr)
                else:
                    newstr = re.sub(r'([^%])%'+c,r'\1{:%c}'%c,newstr)
            newstr.replace('%%','%')

            #newstr = re.sub(f'^URHO3D_LOG{level}F',f'U3D_LOG{level}F',newstr)
            newstr = re.sub(f'^URHO3D_LOG{level}F',f'URHO3D_LOG{level}',newstr)
            print(newstr)
            content = content.replace(m.group(),newstr)
            assert(m.group() not in content)
    return content

files = []
[files.extend(glob(f'../../Source/{d}/**/*.{ext}',recursive=True)) for ext in ['h','cpp'] for d in ['Samples','Urho3D','Tools']]
for fname in files:
    #print(fname)
    with open(fname,'r') as f: # '../../Source/Urho3D/Graphics/OpenGL/OGLGraphics.cpp'
        c = convert(f)
    if c is not None:
        with open(fname,'w') as f: # '../../Source/Urho3D/Graphics/OpenGL/OGLGraphics.cpp'
            f.write(c)

print('\n\nFINISHED\n\n')
if bad:
    print('Failed on \n' + '\n'.join(bad))

[SirNate0]

To fix the linker errors in the shared builds, I am going to switch to FMT_HEADER_ONLY.

[SirNate0]

@eugeneko I saw that RBFX is using CMake's target_compile_definitions to specify FMT_SHARED/FMT_EXPORT in building fmt. Is there a reason for taking this approach over always using the HEADER_ONLY form?

[eugeneko]

Slightly faster compilation, probably? I don't think it matters much.