Release referenced resources in the destructor for ESP8266SSDP

[mcbp223]

I need this to resolve a crash when I dynamically turn WiFi off, with the SSDP destructor called before WiFi.disconnect().

[d-a-v]

Regarding those, I am using path shortening in one of my projects
let me know if you want me to make a PR with this change in debug.h.

Yes !

edit: Well, @earlephilhower would have to comment here too

[mcbp223]

@d-a-v Now I like more the solution with relative paths, or anything with gcc support as commented later in the thread, since what I've been using is C++ only, and parameter pack metaprogramming hurts your eyes. But for tests, with the following code I get no more /home occurrences inside the .ino.bin file.

#include "Indexes.h"

#ifdef __cplusplus

template <size_t N>
inline constexpr int lastIndexOfSep(const char (&str)[N], int crt) {
	return crt < 0 ? -1
	               : str[crt] == '/' || str[crt] == '\\'
	                   ? crt
	                   : lastIndexOfSep(str, crt - 1);
}

template <size_t N>
inline constexpr int lastIndexOfSep(const char (&str)[N]) {
	return lastIndexOfSep(str, N - 2); // allow for one character filenames
}

template <size_t N_LONG, int LAST_SEP
        , typename Indexes = typename andrivet::ADVobfuscator::Make_Indexes<N_LONG - (LAST_SEP + 1)>::type>
struct FileNameShortener;

  // Partial specialization, used for indices argument deduction
template<size_t N_LONG, int LAST_SEP, int... I>
struct FileNameShortener<N_LONG, LAST_SEP, andrivet::ADVobfuscator::Indexes<I...>> {
	using storage_type = char[sizeof...(I)];
	storage_type buffer_;

	constexpr FileNameShortener(const char (&str)[N_LONG])
		: buffer_ { str[I + (LAST_SEP + 1)]... } {}
};

#ifndef PROGMEM_G
// per solution in https://github.com/esp8266/Arduino/issues/3369
#define PROGMEM_G __attribute__((section(".irom.text.group")))
#endif

#define __SHORT_FILE__ (__extension__({static constexpr FileNameShortener<sizeof(__FILE__), lastIndexOfSep(__FILE__)> PROGMEM_G sf__(__FILE__); FPSTR(sf__.buffer_);}))
#else // not C++
#define __SHORT_FILE__ (__builtin_strrchr(__FILE__, '/') ? __builtin_strrchr(__FILE__, '/') + 1 : __FILE__)
//#define __SHORT_FILE__ __PRETTY_FUNCTION__
//#define __SHORT_FILE__ __FUNCTION__
#endif

[d-a-v]

and parameter pack metaprogramming hurts your eyes

Show me where I wrote that

I like your snippet. While c++ constexpr functions are elegant, are they really needed regarding the simplicity of the c-only define which is also c++ compatible ?

Now I like more the solution with relative paths

It seems easier to replace __FILE__ by __SHORT_FILE__ than to patch gcc no ?
(and according to @earlephilhower these patches are rather hairy and now unmaintained)

[mcbp223]

While c++ constexpr functions are elegant, are they really needed regarding the simplicity of the c-only define which is also c++ compatible

As far as I understand, the name of the game here is to convince the compiler to put in the binary the short file name only, for several reasons - to reduce memory consumption, privacy purposes, be able to produce the same exact binary on different machines with different path arrangements. To cut the filename short you'd have to execute a compile-time processing, and constexpr is designed to require the compiler to do a computation at compile-time, instead of at run-time. Also, without the int... I parameter pack gymnastics I wasn't able to completely remove the full filename from the output, no matter what else I tried. Too bad I can't use this in the C parts of the program though.

In the solution I've linked to, @Eszartek didn't modify gcc, but only a config file in his development chain. You seemed to like his method too, in a comment you've made in the thread.

[d-a-v]

Ah! Of course you are right.
I don't know why I was thinking __builtin_strrchr was called at compile time...

[earlephilhower]

@d-a-v, It's a 1-line change to convert GCC to only use the filename everywhere __FILE__ is used while leaving debug info untouched as far as I can see.

<removed diff which doesn't seem to actually do anything>

The uncommitted GCC patches allow much more granular transformations, not just eliminations, of multiple paths.

This is a neat discussion but not really related to SSDP. Maybe jump to #4520 if you want to get into it, for everyone's sanity?

[earlephilhower]

@d-a-v, It's a 1-line change to convert GCC to only use the filename everywhere __FILE__ is used while leaving debug info untouched as far as I can see.
....

After actually trying this patch I can see the lines changed don't actually get called on __FILE__ expansion somehow. Too bad, it looked so simple and clean (famous last words). 😞