NOTE: Please see http://bugs.cheetahtemplate.org for future feature requests/bugs/TODO =============================================================================== =============================================================================== Desired for Cheetah 2.0 ======================= - Smart HTML filter that escapes all values except those individually marked as preformatted, a la Kid/PTL/QPY. (MO) TODO Items (many are just ideas. This is not an official roadmap!) ================================================================================ - "cheetah test" problem: subcommands fail mysteriously on Windows. Rewrite to avoid using subcommands. Instead, set sys.argv and call the appropriate main() for each test. - Documentation: document #encoding. Explain problems "cheetah test" if they haven't been fixed yet. - There's a kludge in CheetahWrapper.py to abort with a helpful error message if the user runs 'cheetah test' but doesn't have write permission in the current directory. The tests should instead put their temporary files under the system tmp directory. - Reset the current filter to the default (or to the constructor's filter if specified) at the beginning of each fill. Currently, filter changes leak from one fill to the next. - CheetahWrapper stuff: (MO) * "cheetah preview [options] [FILES]" print template-specific portion of main method(s) to stdout, with line numbers based on the .py template module. Make a Template method to do the same thing, a la .generatedModuleCode(). * Refactor, make compile/fill/code routines callbacks using a bundle arg. * If an input file ends in a dot, intelligently add the input extension if not found. - Debugging tools. See section below. - Provide a utility to list the names of all placeholders in the template. Requested by Tracy Ruggles on Feb 21, 2003. - 'errorCatcher None' to stop catching errors in the middle of a template. - Utils.WebInputMixin: factor out Cheetah-specific code so it can be used in non-Cheetah applications. Don't modify the searchList: have a Template wrapper method do that. Consider turning it into a function that does not require 'self'. Consider making Webware-specific code into plugins so that, e.g., other cookie-handling methods can be grafted in. Maybe use callback classes like the planned rewrite for CheetahWrapper. Low priority. (MO) - Look through Zope Page Templates (TAL) for ideas to borrow. http://www.zope.org/Documentation/Books/ZopeBook/current/AppendixC.stx http://www.owlfish.com/software/simpleTAL/index.html Debugging Tools (Dump Tools) ============================ It would be nice to provide debugging tools for users who can't figure out why a certain placeholder value isn't found or is being overridden. My idea is to define $dumpSearchList() and $dumpSearchListFlat() in Template, which would print a stanza in the output showing all searchList variables and their values. $dumpSearchList would group by searchList element; $dumpSearchListFlat would combine all into a single alphabetical listing. I made an experimental version but it printed only instance variables, not methods and not inherited attributes. Also, it wouldn't print right using the usual pattern of write-to-custom-StringIO-object-and-return- the-.getvalue() and I couldn't figure out why. The stanza should be set apart by a row of stars with the words "BEGIN/END SEARCH LIST DUMP". Then for $dumpSearchList, precede each group with "*** searchList[i], type , 142 variables ***". Because some elements like 'self' may have hundreds of inherited methods that would create a forest-through-trees situation for the user, we may need an option to supress the variable listing for elements with > 20 variables (just print the summary line instead). ? The final version should be in Template so it has implicit access to the searchList and perhaps later to other variables (locals, globals, "#set global"s, builtins) too. This is such a central debugging tool that you should not have to monopolize an #extends (the template's only #extends) to use it. You could import it, however, if you pass in the searchList explicitly as an argument. In that case, perhaps we can base it on a generic module for dumping variables/values. Note that we cannot simply depend on str() and pprint, because we need to show instances as dictionaries. Likewise, dir() and vars() may get us part of the distance, but only if they show methods and inherited attributes too. These functions should print only top-level variables, not the subelements of collections. I.e, if the first searchList element is a dictionary, show its keys/values, but do not expand any subvalues if they are dictionaries too, unless the display tool happens to default to that. #entry $func($arg1, $arg2="default", $**kw) =============================================================================== Make a wrapper function in the .py template module that builds a searchList from its positional arguments, then instantiates and fills a template and returns the result. The preceding example would create a function thus: def func(arg1, arg2="default", searchList=None, **kw): """Function docstring.""" sl = {'arg1': arg1, 'arg2': arg2} if searchList is None: searchList = [sl] elif type(searchList) == types.ListType: searchList.insert(0, sl) else: raise TypeError("arg 'searchList'") t = TheTemplate(searchList=searchList, **kw) return str(t) ##doc-entry: and #*doc-entry: comments are appended to the function docstring. Finally, make this function accessible directly from the shell. If there are any non-option arguments on the command line, call the function instead of filling the template the normal way. This would perhaps make more sense as arguments to .respond(). But .respond() has that pesky 'trans' argument that mustn't be interfered with, and other programs may assume .respond() takes only one argument. Also, when called via str(), str() cannot take arguments. #indent ======================================================================== The current indenter (which exists but is undocumented) is a kludge that has an indentation object, with implicit placeholder calls added at each line to generate the indentation, and #silent calls to adjust the object. It should be reimplemented to generate code to call the indentation object directly. Also, the user interface should be cleaned up, the implementation and Users' Guide synchronized, and test cases built. The desired implementation revolves around self._indenter, which knows the current indentation level (a non-negative integer), chars (the string output per level, default four spaces), and stack (the previous indentation levels). The .indent() method returns the indentation string currently appropriate. The desired interface for phase 1 (subject to change): #indent strip ; strip leading whitespace from input lines #indent add ; add indentation to output lines as appropriate #indent on ; do both #indent off ; do neither #indent reset ; set level to 0 and clear stack #indent ++ ; increment level #indent -- ; decrement level #indent pop [EXPR] ; revert to Nth previous level (default 1) ; if pop past end of stack, set level to 0 and ; clear stack. All +/-/= operations push the old level ; onto the stack. #indent debug ; dump level, chars and stack to template output Possible extensions: #indent =EXPR ; set level to N (likely to be added to phase 1) #indent +EXPR ; add N to level (not very necessary) #indent -EXPR ; subtract N from level (not very necessary) #indent balance BOOL ; require all indent changes in a #def/#block to be ; popped before exiting the method. (difficult to ; implement) #indent implicitPop BOOL ; automatically pop indent changes within a ; #def/block when that method exits. (difficult to ; implement) #indent ?? ; a 3-way switch that combines unbalanced, balanced and ; implicit pop. (difficult to implement) #indent ?? ; smart stripping: strip input indentation according to ; nested directive level; e.g., ; 01: #if foo=1 ; 02: public int foo() ; 03: { ; 04: return FOO; ; 05: } ; 06: #end if ; With smart stripping, line 4 would be indented and the ; others not. With "on" or "strip" stripping, all lines ; 2-5 would be unindented. With "off" stripping, ; lines 2-5 would not be stripped. There should be one indentation object per Template instance, shared by methods and include files. Upload File ======================================================================== @@TR: This is way outside Cheetah's scope! A mixin method in Cheetah.Utils (for Template) that handles file uploads -- these are too complicated for .webInput(). The method should do a "safe" file upload; e.g., http://us3.php.net/manual/en/features.file-upload.php , within the limitations of Python's cgi module. The user has the choice of three destinations for the file contents: (A) copied to a local path you specify, (B) placed in a namespace variable like .cgiImport() does, or (C) returned. (B) parallels .webInput, but (A) will certainly be desirable situations where we just want to save the file, not read it into memory. Reject files larger than a user-specified size or not in a list of user-approved MIME types. Define appropriate exceptions for typical file-upload errors. Method name .webUploadFileAsString? One situation to support is when form has a text(area) field related to a file-upload control on the same form, and the user has the choice of typing into the field or uploading a text file. We need a method that updates the text field's value if there is an uploaded file, but not if there isn't. This may be handled by the regular method(s) or may require a separate method. RPM Building ============ From: John Landahl To: cheetahtemplate-discuss@lists.sourceforge.net Subject: [Cheetahtemplate-discuss] Building Cheetah RPMs Date: Wed, 05 Nov 2003 01:27:24 -0800 If anyone is interested in building Cheetah RPMs, simply add the following lines to a file called MANIFEST.in in the Cheetah directory and you'll be able to use the "bdist_rpm" option to setup.py (i.e. "python setup.py bdist_rpm"): include SetupTools.py include SetupConfig.py include bin/* Also, I've found that using /usr/lib/site-python for add-on Python packages is much more convenient than the default of /usr/lib/pythonX/site-packages, especially when jumping back and forth between 2.2 and 2.3. If you'd like Cheetah in /usr/lib/site-python, createa a setup.cfg with the following contents: [install] install-lib = /usr/lib/site-python Of course if you do have version specific libraries they should stay in /usr/lib/pythonX/site-packages, but Cheetah seems happy in both 2.2 and 2.3 and so is a good candidate for /usr/lib/site-python. User-defined directives ======================================================================= IF we decide to support user-defined directives someday, consider Spyce's interface. Spyce uses a base class which provides generic services to custom "active tags". http://spyce.sourceforge.net/doc-tag.html http://spyce.sourceforge.net/doc-tag_new.html Test Suite ================================================================================ - add cases that test the cheetah-compile script - add cases that test the integration with various webdev frameworks Examples ================================================================================ - create some non-html code generation examples - SQL - LaTeX - form email - Template definitions in a database. .py template modules in a database? Caching template classes and/or instances extracted from a database. - Pickled templates?