Modifying The TIFF Library

This documentation is best viewed using a graphical browser that supports the latest HTML directives for formatting documents. In particular, this document was authored for viewing with version 1.1 or newer of the Netscape Navigator.


This chapter provides information about the internal structure of the library, how to control the configuration when building it, and how to add new support to the library. The following sections are found in this chapter:


Library Configuration

Information on compiling the library is given elsewhere in this documentation. This section describes the low-level mechanisms used to control the optional parts of the library that are configured at build time. Control is based on a collection of C defines that are specified either on the compiler command line or in a configuration file such as port.h (as generated by the configure script for UNIX systems) or tiffconf.h.

Configuration defines are split into three areas:

If the define COMPRESSION_SUPPORT is not defined then a default set of compression schemes is automatically configured: CCITT Group 3 and 4 algorithms (compression codes 2, 3, 4, and 32771), the Macintosh PackBits algorithm (compression 32773), a Lempel-Ziv & Welch (LZW) algorithm (compression 5), a 4-bit run-length encoding scheme from ThunderScan (compression 32809), and a 2-bit encoding scheme used by NeXT (compression 32766). To override this behaviour define COMPRESSION_SUPPORT and then one or more additional defines to enable configuration of the appropriate codecs (see the table below); e.g.

Several other compression schemes are configured separately from the default set because they depend on ancillary software packages that are not distributed with libtiff.

Support for JPEG compression is controlled by JPEG_SUPPORT. The JPEG codec that comes with libtiff is designed for use with release 5 or later of the Independent JPEG Group's freely available software distribution. This software can be retrieved from the directory ftp.uu.net:/graphics/jpeg/.

NOTE: Enabling JPEG support automatically enables support for the TIFF 6.0 colorimetry and YCbCr-related tags.

Experimental support for the deflate algorithm is controlled by DEFLATE_SUPPORT. The deflate codec that comes with libtiff is designed for use with version 0.99 or later of the freely available libz library written by Jean-loup Gailly and Mark Adler. The data format used by this library is described in the files zlib-3.1.doc, and deflate-1.1.doc, available in the directory ftp.uu.net:/pub/archiving/zip/doc. The library can be retried from the directory ftp.uu.net:/pub/archiving/zip/zlib/ (or try quest.jpl.nasa.gov:/beta/zlib/).

NOTE: The deflate algorithm is experimental. Do not expect to exchange files using this compression scheme; it is included only because the similar, and more common, LZW algorithm is claimed to be governed by licensing restrictions.

By default tiffconf.h defines COLORIMETRY_SUPPORT, YCBCR_SUPPORT, and CMYK_SUPPORT.

DefineDescription
CCITT_SUPPORT CCITT Group 3 and 4 algorithms (compression codes 2, 3, 4, and 32771)
PACKBITS_SUPPORT Macintosh PackBits algorithm (compression 32773)
LZW_SUPPORT Lempel-Ziv & Welch (LZW) algorithm (compression 5)
THUNDER_SUPPORT 4-bit run-length encoding scheme from ThunderScan (compression 32809)
NEXT_SUPPORT 2-bit encoding scheme used by NeXT (compression 32766)
OJPEG_SUPPORT obsolete JPEG scheme defined in the 6.0 spec (compression 6)
JPEG_SUPPORT current JPEG scheme defined in TTN2 (compression 7)
ZIP_SUPPORT experimental Deflate scheme (compression 32946)
PIXARLOG_SUPPORT Pixar's compression scheme for high-resolution color images (compression 32909)
COLORIMETRY_SUPPORT support for the TIFF 6.0 colorimetry tags
YCBCR_SUPPORT support for the TIFF 6.0 YCbCr-related tags
CMYK_SUPPORT support for the TIFF 6.0 CMYK-related tags


General Portability Comments

This software is developed on Silicon Graphics UNIX systems (big-endian, MIPS CPU, 32-bit ints, IEEE floating point). The configure shell script generates the appropriate include files and make files for UNIX systems. Makefiles exist for non-UNIX platforms that the code runs on -- this work has mostly been done by other people.

In general, the code is guaranteed to work only on SGI machines. In practice it is highly portable to any 32-bit or 64-bit system and much work has been done to insure portability to 16-bit systems. If you encounter portability problems please return fixes so that future distributions can be improved.

The software is written to assume an ANSI C compilation environment. If your compiler does not support ANSI function prototypes, const, and <stdarg.h> then you will have to make modifications to the software. In the past I have tried to support compilers without const and systems without <stdarg.h>, but I am no longer interested in these antiquated environments. With the general availability of the freely available GCC compiler, I see no reason to incorporate modifications to the software for these purposes.

An effort has been made to isolate as many of the operating system-dependencies as possible in two files: tiffcomp.h and libtiff/tif_<os>.c. The latter file contains operating system-specific routines to do I/O and I/O-related operations. The UNIX (tif_unix.c), Macintosh (tif_apple.c), and VMS (tif_vms.c) code has had the most use; the MS/DOS support (tif_msdos.c) assumes some level of UNIX system call emulation (i.e. open, read, write, fstat, malloc, free).

Native CPU byte order is determined on the fly by the library and does not need to be specified. The HOST_FILLORDER and HOST_BIGENDIAN definitions are not currently used, but may be employed by codecs for optimization purposes.

The following defines control general portability:

BSDTYPES Define this if your system does NOT define the usual BSD typedefs: u_char, u_short, u_int, u_long.
HAVE_IEEEFP Define this as 0 or 1 according to the floating point format suported by the machine. If your machine does not support IEEE floating point then you will need to add support to tif_machdep.c to convert between the native format and IEEE format.
HAVE_MMAP Define this if there is mmap-style support for mapping files into memory (used only to read data).
HOST_FILLORDER Define the native CPU bit order: one of FILLORDER_MSB2LSB or FILLORDER_LSB2MSB
HOST_BIGENDIAN Define the native CPU byte order: 1 if big-endian (Motorola) or 0 if little-endian (Intel); this may be used in codecs to optimize code

On UNIX systems HAVE_MMAP is defined through the running of the configure script; otherwise support for memory-mapped files is disabled. Note that tiffcomp.h defines HAVE_IEEEFP to be 1 (BSDTYPES is not defined).


Types and Portability

The software makes extensive use of C typedefs to promote portability. Two sets of typedefs are used, one for communication with clients of the library and one for internal data structures and parsing of the TIFF format. There are interactions between these two to be careful of, but for the most part you should be able to deal with portability purely by fiddling with the following machine-dependent typedefs:

uint8 8-bit unsigned integer tiff.h
int8 8-bit signed integer tiff.h
uint16 16-bit unsigned integer tiff.h
int16 16-bit signed integer tiff.h
uint32 32-bit unsigned integer tiff.h
int32 32-bit signed integer tiff.h
dblparam_t promoted type for floats tiffcomp.h

(to clarify dblparam_t, it is the type that float parameters are promoted to when passed by value in a function call.)

The following typedefs are used throughout the library and interfaces to refer to certain objects whose size is dependent on the TIFF image structure:

typedef unsigned int ttag_t; directory tag
typedef uint16 tdir_t; directory index
typedef uint16 tsample_t; sample number
typedef uint32 tstrip_t; strip number
typedef uint32 ttile_t; tile number
typedef int32 tsize_t; i/o size in bytes
typedef void* tdata_t; image data ref
typedef void* thandle_t; client data handle
typedef int32 toff_t; file offset (should be off_t)
typedef unsigned char* tidata_t; internal image data

Note that tstrip_t, ttile_t, and tsize_t are constrained to be no more than 32-bit quantities by 32-bit fields they are stored in in the TIFF image. Likewise tsample_t is limited by the 16-bit field used to store the SamplesPerPixel tag. tdir_t constrains the maximum number of IFDs that may appear in an image and may be an arbitrary size (without penalty). ttag_t must be either int, unsigned int, pointer, or double because the library uses a varargs interface and ANSI C restricts the type of the parameter before an ellipsis to be a promoted type. toff_t is defined as int32 because TIFF file offsets are (unsigned) 32-bit quantities. A signed value is used because some interfaces return -1 on error (sigh). Finally, note that tidata_t is used internally to the library to manipulate internal data. User-specified data references are passed as opaque handles and only cast at the lowest layers where their type is presumed.


General Comments

The library is designed to hide as much of the details of TIFF as possible. In particular, TIFF directories are read in their entirety into an internal format. Only the tags known by the library are available to a user and certain tag data may be maintained that a user does not care about (e.g. transfer function tables).


Adding New Tags

To add support for a new directory tag you have three options. If your tag is specific to a compression algorithm, see below. If you have a lot of tags you may want to try using Niles Ritter's runtime tag-extension scheme in the "contrib/tags" directory, which makes the changes orthogonal to the main libtiff code. Otherwise use the following guidelines to add support to the ``core library''.
  1. Define the tag in tiff.h.
  2. Add a field to the directory structure in tif_dir.h and define a FIELD_* bit.
  3. Add an entry in the TIFFFieldInfo array defined at the top of tif_dirinfo.c. Note that you must keep this array sorted by tag number and that the widest variant entry for a tag should come first (e.g. LONG before SHORT).
  4. Add entries in _TIFFVSetField() and _TIFFVGetField() for the new tag.
  5. (optional) If the value associated with the tag is not a scalar value (e.g. the array for TransferFunction) and requires special processing, then add the appropriate code to TIFFReadDirectory() and TIFFWriteDirectory(). You're best off finding a similar tag and cribbing code.
  6. Add support to TIFFPrintDirectory() in tif_print.c to print the tag's value.

If you want to maintain portability, beware of making assumptions about data types. Use the typedefs (uint16, etc. when dealing with data on disk and t*_t when stuff is in memory) and be careful about passing items through printf or similar vararg interfaces.


Adding New Builtin Codecs

To add builtin support for a new compression algorithm, you can either use the "tag-extension" trick to override the handling of the TIFF Compression tag (see Adding New Tags, above), or do the following to add support directly to the core library:
  1. Define the tag value in tiff.h.
  2. Edit the file tif_codec.c to add an entry to the _TIFFBuiltinCODECS array (see how other algorithms are handled).
  3. Add the appropriate function prototype declaration to tiffiop.h (close to the bottom).
  4. Create a file with the compression scheme code, by convention files are named tif_*.c (except perhaps on some systems where the tif_ prefix pushes some filenames over 14 chars.
  5. Edit Makefile.in (and any other Makefiles) to include the new source file.

A codec, say foo, can have many different entry points:

TIFFInitfoo(tif, scheme)/* initialize scheme and setup entry points in tif */
fooSetupDecode(tif)	/* called once per IFD after tags has been frozen */
fooPreDecode(tif, sample)/* called once per strip/tile, after data is read,
			    but before the first row is decoded */
fooDecode*(tif, bp, cc, sample)/* decode cc bytes of data into the buffer */
    fooDecodeRow(...)	/* called to decode a single scanline */
    fooDecodeStrip(...)	/* called to decode an entire strip */
    fooDecodeTile(...)	/* called to decode an entire tile */
fooSetupEncode(tif)	/* called once per IFD after tags has been frozen */
fooPreEncode(tif, sample)/* called once per strip/tile, before the first row in
			    a strip/tile is encoded */
fooEncode*(tif, bp, cc, sample)/* encode cc bytes of user data (bp) */
    fooEncodeRow(...)	/* called to decode a single scanline */
    fooEncodeStrip(...)	/* called to decode an entire strip */
    fooEncodeTile(...)	/* called to decode an entire tile */
fooPostEncode(tif)	/* called once per strip/tile, just before data is written */
fooSeek(tif, row)	/* seek forwards row scanlines from the beginning
			   of a strip (row will always be >0 and <rows/strip */
fooCleanup(tif)		/* called when compression scheme is replaced by user */

Note that the encoding and decoding variants are only needed when a compression algorithm is dependent on the structure of the data. For example, Group 3 2D encoding and decoding maintains a reference scanline. The sample parameter identifies which sample is to be encoded or decoded if the image is organized with PlanarConfig=2 (separate planes). This is important for algorithms such as JPEG. If PlanarConfig=1 (interleaved), then sample will always be 0.


Adding New Codec-private Tags

To add tags that are meaningful only when a particular compression algorithm is used follow these steps:
  1. Define the tag in tiff.h.
  2. Allocate storage for the tag values in the private state block of the codec.
  3. Insure the state block is created when the codec is initialized.
  4. At TIFFInitfoo time override the method pointers in the TIFF structure for getting, setting and printing tag values. For example,
        sp->vgetparent = tif->tif_vgetfield;
        tif->tif_vgetfield = fooVGetField;	/* hook for codec tags */
        sp->vsetparent = tif->tif_vsetfield;
        tif->tif_vsetfield = fooVSetField;	/* hook for codec tags */
        tif->tif_printdir = fooPrintDir;	/* hook for codec tags */
    
    (Actually you may decide not to override the tif_printdir method, but rather just specify it).
  5. Create a private TIFFFieldInfo array for your tags and merge them into the core tags at initialization time using _TIFFMergeFieldInfo; e.g.
        _TIFFMergeFieldInfo(tif, fooFieldInfo, N(fooFieldInfo));
    
    (where N is a macro used liberaly throughout the distributed code).
  6. Fill in the get and set routines. Be sure to call the parent method for tags that you are not handled directly. Also be sure to set the FIELD_* bits for tags that are to be written to the file. Note that you can create ``pseudo-tags'' by defining tags that are processed exclusively in the get/set routines and never written to file (see the handling of TIFFTAG_FAXMODE in tif_fax3.c for an example of this).
  7. Fill in the print routine, if appropriate.
Note that space has been allocated in the FIELD_* bit space for codec-private tags. Define your bits as FIELD_CODEC+<offset> to keep them away from the core tags. If you need more tags than there is room for, just increase FIELD_SETLONGS at the top of tiffiop.h.


Other Comments

The library handles most I/O buffering. There are two data buffers when decoding data: a raw data buffer that holds all the data in a strip, and a user-supplied scanline buffer that compression schemes place decoded data into. When encoding data the data in the user-supplied scanline buffer is encoded into the raw data buffer (from where it is written). Decoding routines should never have to explicitly read data -- a full strip/tile's worth of raw data is read and scanlines never cross strip boundaries. Encoding routines must be cognizant of the raw data buffer size and call TIFFFlushData1() when necessary. Note that any pending data is automatically flushed when a new strip/tile is started, so there's no need do that in the tif_postencode routine (if one exists). Bit order is automatically handled by the library when a raw strip or tile is filled. If the decoded samples are interpreted by the decoding routine before they are passed back to the user, then the decoding logic must handle byte-swapping by overriding the tif_postdecode routine (set it to TIFFNoPostDecode) and doing the required work internally. For an example of doing this look at the horizontal differencing code in the routines in tif_predict.c.

The variables tif_rawcc, tif_rawdata, and tif_rawcp in a TIFF structure are associated with the raw data buffer. tif_rawcc must be non-zero for the library to automatically flush data. The variable tif_scanlinesize is the size a user's scanline buffer should be. The variable tif_tilesize is the size of a tile for tiled images. This should not normally be used by compression routines, except where it relates to the compression algorithm. That is, the cc parameter to the tif_decode* and tif_encode* routines should be used in terminating decompression/compression. This ensures these routines can be used, for example, to decode/encode entire strips of data.

In general, if you have a new compression algorithm to add, work from the code for an existing routine. In particular, tif_dumpmode.c has the trivial code for the "nil" compression scheme, tif_packbits.c is a simple byte-oriented scheme that has to watch out for buffer boundaries, and tif_lzw.c has the LZW scheme that has the most complexity -- it tracks the buffer boundary at a bit level. Of course, using a private compression scheme (or private tags) limits the portability of your TIFF files.


Sam Leffler / sam@engr.sgi.com. Last updated: $Date: 1996/04/29 19:18:06 $