In stb_write_image, function have a 4° parameter: stride_in_bytes

int stbi_write_png(char const *filename, int w, int h, int comp, const void *data, int stride_in_bytes);

so writing

unsigned char* buffer = (unsigned char_)malloc (width * height * 3); glReadPixels (0, 0, width, height, GL_RGB, GL_UNSIGNED_BYTE, buffer); unsigned char_ last_row = buffer + (width * 3 * (height - 1)); stbi_write_png (name, width, height, 3, last_row, -3 * width)

There is no need to flip the image OpenGL.

could it have on tje_encode_to_file_at_quality a fourth identical parameter?

With use of function tje_encode_to_file_at_quality:

glReadPixels( (rw-rh)/2, 0, rh, rh, GL_RGB, GL_UNSIGNED_BYTE, pixels);
tje_encode_to_file_at_quality(filename.c_str(), 3,rh, rh, 3, pixels);

The Gimp say me: Corrupt JPEG data: 26 extraneous bytes before marker 0xdb

I'm sorry to ask for that again but with the time that has passed, I'm unable to manipulate your source code to save an image from the OpenGL buffer to avoid symmetry of the image.

You gave me an indication but I forgot!

The problem is that in OpenGL I get this image while I should get the image by vertical mirror.

Thanks again for everything. Sorry to ask the same question twice!

Dear Sergio,

Thanks for the library, I appreciate the small one header library. I really like it. Because I use the library in a multithreaded program on FreeBSD I made some necessary changes. The library can now compile correctly on FreeBSD. The global buffer is removed and added to the state so no race conditions occur. Will you consider importing these changes into master? I more changes are necessary, I would like to help.

Thanks in advance, Bernard van Gastel

Previously, this assumed that native endianness was always little-endian, which is true for x86 and most other architectures, but untrue for some (mostly older) unusual architectures.

with bug comment fixed:

Function tjei_encode_main crash at execution

When compile, GCC say:

In function ‘void* memcpy(void_, const void_, size_t)’, inlined from ‘int tjei_encode_main(TJEState_, const unsigned char_, int, int, int)’ at tiny_jpeg.h:948:59: /usr/include/x86_64-linux-gnu/bits/string3.h:53:71: warning: call to void* builtin___memcpy_chk(void, const void, long unsigned int, long unsigned int) will always overflow destination buffer return __builtin___memcpy_chk (__dest, __src, __len, __bos0 (__dest));

with build commit /f6d01b81294e10b5ccf7700918eef60eeb16e8f1

g++ say: (with params -Wall -Wextra)

tiny_jpeg.h: In function ‘int tje_encode_to_file_at_quality(const char_, int, int, int, int, const unsigned char_)’:

tiny_jpeg.h:1130:26: warning: missing initializer for member ‘TJEState_s::ehuffcode’ tiny_jpeg.h:1130:26: warning: missing initializer for member ‘TJEState_s::ht_bits’ tiny_jpeg.h:1130:26: warning: missing initializer for member ‘TJEState_s::ht_vals’ tiny_jpeg.h:1130:26: warning: missing initializer for member ‘TJEState_s::qt_luma’ tiny_jpeg.h:1130:26: warning: missing initializer for member ‘TJEState_s::qt_chroma’ tiny_jpeg.h:1130:26: warning: missing initializer for member ‘TJEState_s::fd’

I was using this for a UWP app/library in unity, and UWP/win10 sdk doesn't allow fopen usage (or maybe the unsafe version, I can't remember, I was in a rush, either way, I had to remove these :)

So I needed to make a change to cleanly exclude that support.

This is that change!

Tested on Visual Studio 2015, should work on 2012, 2013, 2017 and maybe even more.

Visual Studio refuse to compile due to the use of fopen(). This pragma disable the warning and let the compilation proceed.

I took the code from LodePNG https://github.com/lvandeve/lodepng/blob/master/lodepng.cpp#L37-L40 That's not a very important PR but may be useful to somes !

Thanks.