CHANGE HISTORY This is version 2.902: * Added a file, README.Win32, with special instructions for users on systems that lack MakeMaker support. * Worked on the regular expressions that are used on the text-based formats, in hopes of further smoothing out the rift between UNIX and non-UNIX systems. * Added use of Cwd. Filenames passed in are prepended with the current dir if they aren't already absolute pathnames. This was to fix a subtle problem where different files with the same relative names (referenced from different directories) would confuse the cache. * Added the image type to the cached data. Previously, only dimensions were cached. Now, a cache-hit returns the same triple returned originally. * Also added the file's modification-time to the cache, as the (growing!) use of Image::Size in Apache mod_perl environments found that if an image is changed, the cache won't notice. Now it does. * Added recognition of thumbnail-format pics from the popular viewer XV. The value of image type for these is the string "XV". * Integrated a patch that makes html_imgsize generate tags that are XHTML- compliant. Same patch also caught a few doc bugs that I'd missed. * Added more to the README file and the pod documentation, particularly dealing with the Apache web server. * Moved the change history to this new file, from the prior location at the end of README. Changes in 2.901: * Use of the Devel::Modlist package showed that my usage of IO::File pulled in a lot of overhead (which I knew, but this meant I couldn't ignore it any longer). Pulled out all IO::File stuff, use the Symbol package instead for anonymous GLOBs. Also, the calling form that allows an open filehandle be passed in is no longer so picky that it be blessed into IO::File. It does have to be a hard reference, though. Symbolic refs still look like filenames. * While doing the above, I eliminated the last of the strict-refs conflicts. Prior to now, the package ran under strict subs and vars, but not refs. It is now fully strict-clean. Changes in 2.9: * Support for BMP format in 2.8 was marred by a byte-ordering problem not found in testing since my Linux and Windows machines are both Intel-based. Sorry. * Added a contributed fix for a recurring problem with GIF87a images that have descriptors similar to GIF89a. * Added a contributed fix to the PPM matching (prior to actual selection for sizing) that was caused by some Win32 versions of Perl not handling "\n" natively. * While at that last one, I made an attempt to fix any MacOS concerns that may be triggered by the same consideration. * Added better error handling to the imgsize script and added a little more to the manual page. Changes in 2.8: * Added support for BMP-format files. CR/LF translation between UNIX and Windows machines should not affect this, but please let me know if there are problems. Changes in 2.7: * Corrected numerous documentation bugs in which I repeatedly referred to the X and Y dimensions in the wrong order. This was pointed out by several people, but I also kept forgetting to attend to it. * The base routine imgsize now returns a null list when called in a void or scalar context. Mainly, I like pre-loading the cache under mod_perl in Apache (so that spawned children inherit the cache) and didn't like the idea of all those calls populating a stack that was then ignored. Changes in 2.6: * 2.5 did not actually deliver the tests for the new TIFF code. Oops. * Found some variations of the formats of XPM and XBM that were not compatible with the regex's being used. Tricky, flexible little devils, especially XPM. Changes in 2.5: * Added support for TIFF images and two test cases. Many thanks to Cloyce Spradling <firstname.lastname@example.org> for both the code patches and the test images. Changes in 2.4: * Small bug in jpegsize, not returning the string 'JPG' as the third element of the returned list. Figures I have a half-documented feature, and someone actually uses it. :-) Changes in 2.3: * Small fix to the regular expression in xpmsize. * Changed syntax of utilizing AutoLoader to conform to changes in upcoming 5.004 release. Changes in 2.2: * Added calls to binmode on filehandles, for the sake of OS flavors such as OS/2, etc. * Cleaned up a new -w warning present in late versions of 5.003 (leading in to 5.004). Changes in 2.1: * Added in changes to the error handling that were meant for 2.0. * Fixed a bug in attr_imgsize caused by the new return format. Changes in 2.0: * No longer uses filenames to determine image format for examination. * Can operate on already-open IO::File objects. * Can operate on in-memory data strings (buffers). * Documentation overhauled. * imgsize() now returns an ignorable third parameter. This value is a three- letter identifier of the image format (using the standard suffixes. Useful when sizing unknown data to also type the image. * Support for the PPM family (PPM, PGM and PBM) of formats. * JPG code extensively re-written and simplified. * PNG code re-written. * Test cases for PNG and PPM added. Test cases also cover the open-file and in-memory capabilities. * Dropped the "optional second argument" thing in the invocation; no one used it, and it was confusing and difficult to explain clearly. * Various typos cleaned up. Changes in 1.2: Two fixes given to me by Bernd Leibing <email@example.com>: * Fixed bug caused by perl quoting and RCS in Size.pm * imgsize now comes as imgsize.PL and is auto-extracted as part of the make, utilizing your system's configuration to create the start-up #! line. Changes in 1.1: * Added a small, simple script called "imgsize" that gives command-line access to the library functionality. * Discovered and fixed bug in sizing JPEGs. Initial version 1.0: * Library model for sizing code from wwwimagesize. By this, I also include the extension-style infrastructure, MakeMaker usage, test suite, etc. * Changed format of returned data to generic (X, Y), added wrappers to format into "WIDTH=x HEIGHT=y" HTML-style and (-WIDTH => x, -HEIGHT => y) CGI-style. * Added cacheing of sizes based on passed-in file name. I have one CGI script that uses a given icon for a bullet list, with 35+ items in the list... * UNDOCUMENTED UNTIL BETTER TESTED: Optional second argument to imgsize can be either scalar or a code ref; if scalar, is pasted in front of file name as a directory component; if it is a code ref, the routine is called with the file name as a single argument and is expected to return the full path name. Kind of clunky right now, easier to just pass in the full path to begin with. * Added the XPM sizing myself, though I have yet to use it (I do test it in the test suite, though).