tpl User Guide

There is no need to understand the internal structure of the tpl image. But for the curious, the image is a strictly defined binary buffer having two sections, a header and the data. The header encodes the length of the image, its format string, endian order and other flags. The data section contains the packed data.

A property of the tpl image is that consecutive images can be written to a stream without requiring any delimiter between them. The reader making use of tpl_gather (or tpl_load in TPL_FD mode) will obtain exactly one tpl image at a time. Therefore tpl images can be used as an IPC message format without any higher-level framing protocol.

A tpl image generated on one kind of CPU will generally be portable to other CPU types when tpl is used properly. This may be a surprise considering that tpl is a binary format. But tpl has been carefully designed to make this work. Each format character has an associated explicitly-sized type. For integer and floating point types, whose "endian" or byte-order convention varies from one CPU to another, tpl automatically and transparently corrects the endian order (if needed) during the unpacking process. Floating point numbers present their own special difficulties. No guarantees are made with regard to floating point portability. That said, because many modern CPU’s use IEEE 754 floating point representation, data is likely to be portable among them.

While a tpl image is a binary entity, you can view any tpl image in XML format using the included tplxml utility, located in the lang/perl directory.

The utility is bidirectional, as shown. The file extension is not important; tplxml inspects its input to see if it’s tpl or XML. You can also pipe data into it instead of giving it a filename. The tplxml utility is slow. Its purpose is two-fold: debugging (manual inspection of the data in a tpl), and interoperability with XML-based programs. The resulting XML is often ten times the size of the original binary tpl image.

There is a Perl module in lang/perl/Tpl.pm. The Perl API is convenient for writing Perl scripts that interoperate with C programs, and need to pass structured data back and forth. It is written in pure Perl.

Other open-source software by the author is listed at http://troydhanson.github.io

You can compile and run the built-in test suite by running:

Tpl can be used directly (instead of as a DLL) by compiling the tpl sources right into your program. To do this, add tpl.c, tpl.h, win/mman.h and win/mmap.c to your program’s source and header files and add the preprocessor definition TPL_NOLIB.

If you like, you can build the DLL yourself using VS2008 or VS2010 (the free Express Edition is sufficient) and perhaps newer versions, though this has not been tested by the author. (As of 2013 we need to restore the solution file to make this easy, that’s currently missing).

You can build it in the traditional Unix method under Cygwin/MinGW using the "configure; make; make install" approach. If the "configure" script is not present in the repository you cloned, generate it by running "bootstrap".

The sizes of data types such as long and double vary by platform. This must be kept in mind because most tpl format characters require a pointer argument to a specific-sized type, listed above. You can use explicit-sized types such as int32_t (defined in inttypes.h) in your program if you find this helpful.

Before explaining all the concepts, it’s illustrative to see how both kinds of arrays are used. Let’s pack the integers 0 through 9 both ways.

Note that the length of the fixed-length array (10) was passed as an argument to tpl_map(). The corresponding unpacking example is listed further below. Now let’s see how we would pack 0-9 as a variable-length array:

Notice how we called tpl_pack in a loop, once for each element 0-9. Again, there is a corresponding unpacking example shown later in the guide. You might also notice that this time, we passed 1 as the final argument to tpl_pack. This is an index number designating which variable-length array we’re packing. In this case, there is only one.

Index numbers identify a particular variable-length array in the format string. Each A(...) in a format string has its own index number. The index numbers are assigned left-to-right starting from 1. Examples:

A multi-dimensional matrix of integers can be packed and unpacked the same way as any fixed-length array.

This single call to tpl_pack packs the entire matrix.

The s format character is only for use with char* types. In the example above, s is a char*. If it had been a char s[14], we would use the format characters c# to pack or unpack it, as a fixed-length character array. (This unpacks the characters "in-place", instead of into a dynamically allocated buffer). Also, a fixed-length buffer described by c# need not be null-terminated.

You can use fixed- or variable-length arrays of strings in tpl. An example of packing a fixed-length two-dimensional array of strings is shown here.

Later, when unpacking these strings, the programmer must remember to free them one by one, after they are no longer needed.

Arrays of structures are the same as simple arrays. Fixed- or variable- length arrays are supported.

The differences between fixed- and variable-length arrays are explained in the Arrays section.

When dealing with nested structures, the outermost structure uses the S format character, and the inner nested structures use the $ format. Only the outermost structure’s address is given to tpl_map.

Structures can nest to any level. Currently tpl does not support fixed-length array suffixes on inner structures. However the outermost structure can have a length suffix even if it contains some nested structures.

It is necessary to pack index number 0 only if the format string contains characters that are not inside an A(...), such as the i in the format string iA(c).

When reading a tpl image from a file or memory (but not from a file descriptor) the size of the file or memory buffer must exactly equal that of the tpl image stored therein. In other words, no excess trailing data beyond the tpl image is permitted. The bit flag TPL_EXCESS_OK can be OR’d with TPL_MEM or TPL_FILE to relax this requirement.

A situation where this flag can be useful is in conjunction with tpl_dump in the TPL_MEM|TPL_PREALLOCD mode. In this example, the program does not concern itself with the actual tpl size as long as LEN is sufficiently large.

It is necessary to unpack index number 0 only if the format string contains characters that are not inside an A(...), such as the i in the format string iA(c).

The format string can be obtained like this:

On success, a copy of the format string is returned. The caller must eventually free it. On error, such as a non-existent file, or an invalid tpl image, it returns NULL.

The lengths of all fixed-length arrays in the format string can be queried using the TPL_FXLENS mode. It provides the number of such fixed-length arrays and their lengths. If the former is non-zero, the caller must free the latter array when finished. The format string itself must also be freed.

The TPL_FXLENS mode is mutually exclusive with TPL_DATAPEEK.

To peek into the data, additional arguments are used. This is a quick alternative to mapping, loading and unpacking the tpl, but peeking is limited to the data in index 0. In other words, no peeking into A(...) types. Suppose the tpl image in file.tpl has the format string siA(i). Then the index 0 format characters are si. This is how to peek at their content:

Now s, i, and fmt have been populated with data. The caller must eventually free fmt and s because they are allocated strings. Of course, it works with TPL_MEM as well as TPL_FILE. Notice that TPL_DATAPEEK was OR’d with the mode. You can also specify any leading portion of the index 0 format if you don’t want to peek at the whole thing:

The TPL_DATAPEEK mode is mutually exclusive with TPL_FXLENS.

The oops has the same prototype as printf. The built-in default oops handling function writes the error message to stderr.

The fatal hook is invoked when a tpl function cannot continue because of an out- of-memory condition or some other usage violation or inconsistency. It has this prototype:

The fatal hook must not return. It must either exit, or if the program needs to handle the failure and keep executing, setjmp and longjmp can be used. The default behavior is to exit(-1).

This example is included in tests/test123.c. When run, this program prints:

In this mode, tpl_gather blocks while reading file descriptor fd until one complete tpl image is read. No bytes past the end of the tpl image will be read. The address of the buffer containing the image is returned in img and its size is placed in sz. The caller is responsible for eventually freeing the buffer. The function returns 1 on success, 0 on end-of-file, or a negative number on error.

This mode is for non-blocking, event-driven programs that implement their own file descriptor readability testing using select() or the like. In this mode, tpl images are gathered in chunks as data becomes readable. Whenever a full tpl image has been gathered, it invokes a caller-specified callback to do something with the image. The arguments are the file descriptor fd which the caller has determined to be readable and which must be in non-blocking mode, a pointer to a file-descriptor-specific handle which the caller has declared (explained below); a callback to invoke when a tpl image has been read; and an opaque pointer that will passed to the callback.

For each file descriptor on which tpl_gather will be used, the caller must declare a tpl_gather_t* and initialize it to NULL. Thereafter it will be used internally by tpl_gather whenever data is readable on the descriptor.

The callback will only be invoked whenever tpl_gather() has accumulated one complete tpl image. It must have this prototype:

The callback can do anything with the tpl image but it must not free it. It can be copied if it needs to survive past the callback’s return. The callback should return 0 under normal circumstances, or a negative number to abort; that is, returning a negative number causes tpl_gather itself to discard any remaining full or partial tpl images that have been read, and to return a negative number (-4 in particular) to signal its caller to close the file descriptor.

The return value of tpl_gather() is negative if an error occured or 0 if a normal EOF was encountered-- both cases require that the caller close the file descriptor (and stop monitoring it for readability, obviously). If the return value is positive, the function succeeded in gathering whatever data was currently readable, which may have been a partial tpl image, or one or more complete images.

This mode is identical to TPL_GATHER_NONBLOCKING except that it gathers from a memory buffer instead of from a file descriptor. In other words, if some other layer of code-- say, a decryption function (that is decrypting fixed-size blocks) produces tpl fragments one-by-one, this mode can be used to reconstitute the tpl images and invoke the callback for each one. Its parameters are the same as for the TPL_GATHER_NONBLOCKING mode except that instead of a file descriptor, it takes a buffer address and size. The return values are also the same as for TPL_GATHER_NONBLOCKING noting of course there is no file descriptor to close on a non-positive return value.