Release notes for Jam/MR [UNRELEASED] (aka Jam - make(1) redux) *** Note: These Release notes represent work in progress, *** not a completed release! 1. Release info: (TENTATIVE) Jam/MR 2.4 (TENTATIVE, IN PROGRESS!) Month, dd, 2002 VERSION 2.4 (TENTATIVE, IN PROGRESS!) 2. Compatibility Jam 2.4 is upward compatible with Jam 2.3 The Jam 2.4 language is a superset of the 2.3 language; Jamfiles, Jambase, and other rulesets used in 2.3 can be used with the 2.4 language support. 3. Changes since 2.3. 3.1. Changes to Jam Language The mechanism for calling rules that return values - "[ rule args ...]", (and 'return' in the rule body), is now a documented part of the language. Add "on <target> <rulename> <field1> ..." syntax, to invoke a rule under the influence of a target's specific variables. Add "[ on targ rule ... ]" to call a rule returning a value, under the influence of a target's specific variables. New 'Glob' builtin that returns a list of files, given a list of directories, and a list of filename patterns. New 'while expr { block }' construct. New :E=value modifier provides default value if variable unset. New :J=joinval modifier concatentates list elements into single element, separated by joinval. \ can now be used to escape a space (or any single whitespace character), so that you don't have to resort to quotes. New 'Match regexp : string' rule matches regexp against string and returns list of results. Rules can now be invoked indirectly, through variable names. If the variable expands to an empty list, no rule is run. If the variable expands to multiple entries, each rule is run with the same arguments. The result of the rule invocation is the concatenation of the results of the rules invoked. 'Echo' and 'Exit' now have aliases 'echo' and 'exit', since it is really hard to tell that these are built-in rules and not part of the language, like 'include'. Real rules continue to start with a capital. 3.2. Jambase Changes Support for YACCGEN, the suffix used on generated yacc output. Fix ups to have jam and p4 build with borland C 5.5, and minor win98 jam support for jam clean SubDirHdrs now takes directory names in the same format as SubInclude : one directory element per word. More portable support for specifying includes and #defines: New ASHDRS, CCHDRS, CCDEFS, DEFINES, ObjectDefines, FQuote, FIncludes, FDefines. Ordering of cc and c++ flags grossly rearranged. Jambase has been compacted by applying the new E: and J: expansion modifiers. 3.3 'jam' Changes (See Jam.html) Added '-q' (quit quick) option; jam will exit promptly (as if it received an interrupt), as soon as any target fails. Added experimental '-g' (build newest sources first) option: all things being equal, normally targets are simply built in the order they appear in the Jamfiles. With this flag, targets with the newest sources are built first. From an idea by Arnt Gulbrandsen. Undocumented (outside this note). 3.4. Jam internal code changes jamgram.yy now defines YYMAXDEPTH to 10000, what it is on FreeBSD, for older yaccs that left it at 150 or so. This is needed for the right-recursion now used in the grammar. Optimize rule compilation, with right-recursion instead of left. Split jam's built-in rules out to builtins.c from compile.c, so that compile.c only deals with the language. Split jam's pathsys.h from filesys.h, since they are really two different pieces. evaluate_if(), which evaluated the condition tree for 'if' and returned an int, has been replaced with compile_eval(), which does essentially the same but returns a LIST. 4. Fixed bugs Missing TEMPORARY targets with multiple parents no longer spoil one parent's time with another. The parents' time is used for comparison with dependents, but no longer taken on as the target's own time. 'actions updated', not 'actions together', now protects targets from being deleted on failed/interrupted updates. Fixed broken $(v[1-]), which always returned an empty expansion. Thanks to Ian Godin <ian@sgrail.com>. Defining a rule within another rule, and invoking the enclosing rule more than once, would result in giving the first rule a null definition. Fixed. $(d:P) now works properly on the mac, climbing up directories. Thanks to Miklos Fazekas <boga@mac.com>. No longer (sometimes) treat \ as a directory separator on UNIX. It isn't supposed to be, but was due to bungled ifdefs. Applying just :U or :D (or :E, :J) mods no longer causes the variable value to be treated as a filename (parsed and rebuilt using the OS specific pathsys routines). Previously, if _any_ mods were present then the value was parsed and rebuilt as if a filename, and that could in certain cases munge the value. Only the file modifiers (:GDBSM) treat the value as a filename. Four rules makeCommon, makeGrist, makeString, makeSubDir from jam 2.2 missing in 2.3 have been readded, with apologies to dtb@cisco.com. Return status more likely to be correct when using -d0, now that targets are could as being built even with no debugging output. Thanks to Miklos Fazekas <boga@mac.com>. yyacc now suffixes all terminals it defines with _t, so that they don't conflict with other symbols (like RULE with the typedef in rules.h). Thanks to Aichael Allard. 5. Porting [MACINTOSH] Paths are now downshifted (interally) so as to handle its case insensitivity. Thanks to Miklos Fazekas <boga@mac.com>. [NT] MS changed the macro for the IA64 Windows NT 64bit compiler. [CYGWIN] Cygwin jam porting: dance around bison and yyacc. Use bison's -y flag to use yacc's output file naming conventions, and don't use yyacc on systems whose SUFEXE is set. [VMS] The Jambase itself was not formatting the CCHDRS and CCDEFS properly: on VMS they can't be appended to, because multiple /define or /include directives don't work. Instead now CCHDRS and CCDEFS is reformatted from HDRS and DEFINES anytime those latter two change. This requires the recent change to jam to allow access to target-specific variables when setting other variables. [VMS] Remove exception call when file_dirscan() can't, for some reason, scan a directory. Use a better set of #ifdefs to determine if we're on a vax, rather than relying on the C compiler being a specific version: we're able to build with the C++ compiler now. [VMS] Port new jam to run with just cxx compiler. (The C compiler being a extra-cost item). [NT] Add entry for DevStudio when the settings are already in the system environment. [NT] default $(MV) to "move /y" in Jambase. =============================================================================== =============================================================================== Release notes for Jam/MR 2.3 (aka Jam - make(1) redux) 0. Bugs fixed since 2.3.1 PATCHLEVEL 2 - 3/12/2001 NOCARE changed back: it once again does not applies to targets with sources and/or actions. In 2.3 it was changed to apply to such targets, but that broke header file builds: files that are #included get marked with NOCARE, but if they have source or actions, they still should get built. 1. Release info: Jam/MR 2.3 November 16, 2000 VERSION 2.3 PATCHLEVEL 1 2. Compatibility Jam 2.3 is upward compatible with Jam 2.2. The Jam 2.3 language is a superset of the 2.2 language; Jamfiles, Jambase, and other rulesets used in 2.2 can be used with the 2.3 language support. 3. Changes since 2.2 3.1. Changes to Jam Language Rules now can have values, which can expanded into a list with the new "[ rule args ... ]" syntax. A rule's value is the value of its last statement, though only the following statements have values: if (value of the leg chosen), switch (ditto), set (value of the resulting variable), return (its arguments). Note that 'return' doesn't actually return. This support is EXPERIEMENTAL and otherwise undocumented. (2.3.1) Because of the new way lists are processed, if a rule has no targets a warning message is no longer issued. NOCARE now applies to targets with sources and/or actions, rather than just those without. 3.2. Jambase Changes The HDRPATTERN variable now allows for leading blanks before the #include, to keep up with ANSI. By john@nanaon-sha.co.jp (John Belmonte) (2.2.3). HDRPATTERN has been adjusted to avoid mistaking cases like: # include <time.h> /* could be <sys/time.h> */ MkDir now NOUPDATE's $(DOT), so that there are no dependencies on the current directory's timestamp. By john@nanaon-sha.co.jp (John Belmonte). The old mock functions like makeDirName, which assigned their results to the variable named as their first argument, have been replaced with real functions using the new [] synxtax. E.g. "makeDirName foo : bar ola" is now "foo = [ fDirName bar ]" Install now always does a cp/chmod/etc, rather than using the system's install(1), which invariably seems broken. 3.3. Jam internal code changes $JAMUNAME is set on UNIX. (2.2.4). Jam ANSI-fied (2.3.0). jam.h now defines a bunch of symbols used by the other source files, so as minimize compiler- and platform-specific ifdefs. OSVER is no longer set by jam.h (it was only set for AIX). Jam does not depend on this variable at all, except to set $(OSFULL), which is used to determine jam's build directory. If the user needs to distinguish between various revs of OSs, he must set OSVER in the environment. 4. Fixed bugs Redefining a rule while it was executing could cause jam to crash. Reference counts are now used to prevent that, thanks to Matt Armstrong. Logic for computing chunk size when executing PIECEMEAL rules has been reworked to be a little more accurate, without danger of overflow, at the cost of being a little more compute intensive. Instead of computing an estimate chunksize in the (now gone) make1chunk(), make1cmds() now just goes full bore and tries to use all args. When that fails, it backs off by 10% of the source args until the command fits. It takes a little bit more compute time compared to the old logic, but when you're executing actions to build all of Shinola it's still pretty small in the scheme of things. The NT handle leak in execunix.c has been fixed, thanks to Gurusamy Sarathy. (2.2.1). 5. Porting Platforms newly supported or updated: AmigaOS (with gcc), courtesy of Alain Penders (2.2.2). Beos CYGWIN 1.1.4, courtesy of John Belmonte <john@nanaon-sha.co.jp>. IBM AS400 via Visual Age on NT (primitive) IBM OS/390 Unix System Services Linux SuSE on OS390 Linux Mips, ARM Lynx HPUX 11, IA64 Mac OS X Server, courtesy of Jeff_Sickel@sickel.com (2.2.5). Mac Rhapsody MPE IX 6.0 NetBSD QNX RTP (QNX 6.0) Siemens Sinix UNICOS VMS 6.2, 7.1 Windows NT IA64 5.1. NT Porting Notes Always create tmp .bat file for actions if JAMSHELL is set. That way, if JAMSHELL is a .bat file itself, it can handle single-command actions with more than 9 cmd line args. COMSPEC is no longer examined: cmd.exe is always used instead. Only cmd.exe can execute the Jambase rules anyhow. Jam can be built with Borland C++ 5.5. OS2 fixes: InstallBin now works. Filenames are now downshifted, so mixed case works better there, too. file_dirscan() can now scan the root ("c:\" or "\") directory, which it couldn't handle before. var_defines now ignores OS=Windows_NT, because it conflicts with Jam's setting of OS (to NT). 5.2. Mac OS 8/9 Notes The support for Mac is curious at best. It runs under MPW. It requires CodeWarrior Pro 5, but no longer requires GUSI. Use Build.mpw to bootstrap the build. The Mac specific definitions in the Jambase are not intended to be of general purpose, but are sufficient to have Jam build itself. =============================================================================== =============================================================================== Release Notes for Jam 2.2 1. Release info: Jam 2.2 October 22, 1997 VERSION 2.2 PATCHLEVEL 1 2. Compatibility Jam 2.2 is a roll-up of 'Jam - make(1) redux' release 2.1+. Most of the changes described below were available before this, in the jam.2.1.plus.tar ball. The Jam 2.2 language is a superset of the 2.1 language; Jamfiles, Jambase, and other rulesets used in 2.1 can be used with the 2.2 language support. See 'Jambase Changes', below, to see if your Jamfiles need any changes to work with the 2.2 Jambase. 3. Changes Since 2.1 New product name: Jam. (Executable program is still named 'jam'.) Documentation rewritten; HTML versions supplied. 3.1 Changes to Jam Language Rules may now have more fields than just $(<) and $(>). Local variables are now supported. The expression 'if $(A) in $(B)' is now supported. New variable modifiers :U and :L result in uppercased or lowercased values. New variable modifier :P reliably results in parent directory of either a file or directory. (Previously, :D was used, but on VMS :D of a directory name is just the directory name.) The :S variable modifier now results in the _last_ suffix if a filename has more than one dot (.) in it. New predefined $(JAMDATE) variable is initialized at runtime for simple date stamping. New predefined variables $(OSVER) and $(OSPLAT) are used to distinguish among operating system versions and hardware platforms, when possible. New 'bind' qualifier on action definitions allows variables other than $(<) and $(>) to be bound with SEARCH and LOCATE paths. Action buffer size is no longer limited by MAXCMD. Instead, each line in an action is limited by MAXLINE, defined for each OS, and the entire action size is limited by CMDBUF. 3.2 Jambase Changes (See Jamfile.html) Jambase has been reworked to incorporate new language features. A handful of new utility rules has been added: makeString, makeDirName, etc. New HDRGRIST variable in Jambase allows for headers with the same name to be distinguished. LOCATE_TARGET now has a new flavor, LOCATE_SOURCE, that is used by rules that generate source files (e.g., Yacc and Lex). Header file includes now happen in the proper order. The limit of 10 include files has been eliminated. The old "Install" rule is no longer available. Use InstallBin, InstallFile, InstallLib, InstallMan, or InstallShell instead. 3.3 'jam' Changes (See Jam.html) 'jam' can now be built as a stand-alone program, with Jambase compiled into the executable. An external or alternate Jambase can still be referenced explicitly with -f. On command failure, 'jam' now emits the text of the command that failed. This is a compromise between the normal -d1 behavior (where commands were never seen) and -d2 (where commands are always seen). 'jam' now exits non-zero if it doesn't have a total success. A parse error, sources that can't be found, and targets that can't be built all generate non-zero exit status. The debugging levels (-d flags) have been slightly redefined. The supplied Jamfile now builds 'jam' into a platform specific subdirectory. This lets you use the same source directory to build 'jam' for more than one platform. The supplied Jamfile does not rebuild generated source files by default. (They are supplied with the distribution.) See Jamfile for more information. 4. Fixed Bugs The 'include' bug has finally been fixed, so that include statements take effect exactly when they are executed, rather than after the current statement block. This also corrects the problem where an 'include' within an 'if' block would wind up including the file one token after the 'if' block's closing brace. Credit goes to Thomas Woods for suggesting that the parse tree generation and parse tree execution be paired in their own loop, rather than having the parser execute the tree directly. The setting and extracting of grist has been regularized: normally, if you set a component of a filename (using the :DBSMG= modifiers), you are supposed to include the delimiters that set off the component: that is, you say "$(x:S=.suffix)", including the ".". But with grist it was inconsistent between setting and getting: setting grist required no <>'s, while getting grist included them. Getting grist continues to return the <>'s, but now setting grist can either include them (the new way) or not (the old way). 'actions together' now suppresses duplicate sources from showing up in $(>). Accessing variables whose names contained ['s (as happens with MkDir on VMS) wasn't working, because it treated the [ as an array subscript. Now [ and ] are, like :, handled specially so that they can appear in variable values. The 'if' statement now compares all elements in expressions; previously, it only compared the first element of each list. If a command line in an action is longer than MAXLINE (formerly MAXCMD), 'jam' now issues an error and exits rather than dumping core. If a Jamfile ended without a trailing newline, jam dumped core. This has been fixed. 5. Porting See jam.h for the definitive list of supported platforms. Since 2.1, support has been added for: Macintosh MPW Alpha VMS Alpha NT NT PowerPC BeOS MVS OE UNIXWARE QNX SINIX (Nixdorf) OS/2 Interactive UNIX (ISC), courtesy of Matthew Newhook 5.1 NT Support Fixes The NT command executor now handles multiple line actions, by writing multi-line actions to a batch file and executing that. Targets are universally lowercased on NT. (Matthew Newhook) Concurrent process support is fully enabled for NT. (Gurusamy Sarathy <gsar@engin.umich.edu>) Path handling: Jam now knows that the directory component of "D:\" is "D:\", just as on unix it knows that the directory component of "/" is "/". It also now successfully gets the timestamp for "D:\" or just plain "\". 5.2 VMS Support Fixes VMS support is much, much better now. The path name manipulation routines (in pathvms.c) were more or less rewritten, and they now handle the vagaries of combining directory and file names properly. Targets are universally lowercased on VMS. Multi-line command blocks on VMS are now executed in a single system() call rather than separate ones for each line, so that actions can be DCL scripts. =============================================================================== =============================================================================== Release notes for Jam 2.1. 1. Release info: Jam 2.1 February 1, 1996 VERSION 2.1 PATCHLEVEL 0 2. Porting Linux is now supported. FREEBSD is now supported. SCO ("M_XENIX") now supported. NCR now supported. NEXT support from karthy@dannug.dk (Karsten Thygesen) DECC support from zinser@axp614.gsi.de (Martin P.J. Zinser) I have changes for OS/2, but no way to test them. Volunteers? I have VMS multiprocess support, but no way to test it. Volunteers? 2.1. NT Support fixes. The NT support is considerably more real than it was in 2.0. Filent.c had its syntax error corrected, it no longer skips the first entry when scanning directories, and it handles string tables in archives (for long object file names). The Jambase was changed a bit to support the various C/C++ compilers on NT, although it has only been thorougly tested with MSVC20. You still need to set MSVCNT or BCCROOT to the root of the the compiler's directory tree, and you'll get an error if you don't set it (rather than getting a pile of mysterious errors). 2.2. Other porting fixes. SPLITPATH now set up for UNIX (:), NT (;), VMS (,) Jambase support for Solaris works better now: the location of AR is hardwired to /usr/ccs/bin/ar and it knowns "install" doesn't take -c. Solaris -- how the mighty have fallen. To handle Linux's wacko yacc, jamgram.h is now included after scan.h so that YYSTYPE is define. 3. Jambase Changes (see Jamfile.html) SubDir now computes the root directory for the source tree, if the variable naming the root directory isn't set in the environment. It counts the number of directory elements leading from the root to the current directory (as passed to SubDir) and uses that many "../"'s to identify the root. This means that to use SubDir you no longer have to have anything special set in the environment. InstallFile is now an alias for InstallLib. 'first' is now dependency of all pseudo-targets (all, files, exe, lib, shell), so that jamming any of these pseudo-targets also builds any dependencies of 'first'. The File rule definition in the Jambase was missing an &. The File rule now calls the Clean rule, so that installed files get cleaned. 4. Jam changes (see Jam.html) Variables may now be set on the command line with -svar=value. Targets marked with NOUPDATE are now immune to the -a (anyhow) flag. Previously, the MkDir rule would try to recreate directories that already exist when jam was invoked with -a. A new variable, $(JAMVERSION), joins the small list of built-in variables. It it set to the release of jam, currently "2.1". If an actions fails, jam now deletes the target(s). It won't delete libraries or other targets that are composites. This is now consistent with jam's behavior on interrupts (it deletes the targets). Jam had a nasty bug when setting multiple variables to the same value: if the first two variable names were the same, the variable value got trashed. This also affected "on target" variables if the first two targets were the same. For example: FOO on bar.c bar.c foo.c = a b c ; This would mangle the value of FOO for bar.c and foo.c. This has been fixed. Jam would generate bogus numbers when reporting the number of targets updated after an interrupt. It now is more careful about counting. The debugging flag -d has been extended. In addition to supporting -dx (turn on debugging for all levels up to x) there is also now -d+x (turn on debugging at only level x). The default output level is -d1 (-or d2 if -n is given); this can be turned off with -d0. The debug levels are listed in jam.1 and jam.h. The parsing debug output now uses indenting to indicate when one rule invokes another. =============================================================================== =============================================================================== Release notes for Jam 2.0. 1. Release info: Jam 2.0 March 10, 1994 VERSION 2.0 PATCHLEVEL 5 2. Porting Windows/NT is now (crudely) supported, courtesy of Brett Taylor and Laura Wingerd. COHERENT/386 is now supported, courtesy of Fred Smith. Solaris archive string table for long archive names is now supported, thanks to Mike Matrigali. 3. Compatibility Jam 2.0 syntax is a superset of Jam 1.0 syntax, and thus it can interpret a Jam 1.0 Jambase. The Jam 2.0 Jambase is a superset of the Jam 1.0 Jambase, and thus it can include a Jamfile written for Jam 1.0. 4. Changes from Jam 1.0 to Jam 2.0 4.1. Documentation changes New Jamfile.5 manual page, with lots of examples and easy reading. It replaces both the old "Examples" file as well as the old Jambase.5 manual page. jam.1 edited by Stephen W. Liddle and Diane Holt. 4.2. Jambase Changes (see Jamfile.5) 4.2.1. New rules: There are new rules to make handling subdirectories easier: SubDir, SubInclude, SubDirCcFlags, SubDirHdrs. There are new rules to handle file-specific CCFLAGS and HDRS: ObjectCcFlags and ObjectHdrs. Misc new rules: HardLink, InstallShell, MkDir. New rule "clean" that deletes exactly what jam has built, and "uninstall" that deletes exactly what was installed. New rules for handling suffixes .s, .f, .cc, .cpp, .C. 4.2.2. Old rules: The InstallBin, Lib, Man, and the new Shell rules now take the destination directory as the target and the files to be copied as sources. These rules formerly took the files to be copied as targets, and used built-in destination directories of $(BINDIR), $(LIBDIR), $(MANDIR), and $(BINDIR). The InstallBin, Lib, Man, and Shell rules use the install(1) program now, instead of doing their own copying. The Cc rule now uses -o when possible, rather than moving the result. Some platforms (Pyramid?) have a broken -o. Jambase rules taking libraries, objects, and executables now all ignore the suffixes provided and use the one defined in the Jambase for the platform. Stupid yyacc support moved out of Jambase, as jam is its only likely user. Jambase now purturbs library sources with a "grist" of SOURCE_GRIST. 4.2.3. Misc: The names of the default rules defined in Jambase have been lowercased and un-abbreviated, to be more imake(1) like. The Jambase has been reorganized and sorted, with VMS and NT support moved in from their own files. The Jambase has been relocated on UNIX from /usr/local/lib/jam to /usr/local/lib. 4.3. Jam changes (see jam.1) 4.3.1. Flags: New -a (anyhow) flag: means build everything. New -j<x> flag: run jobs in parallel. Old -t now rebuilds the touched target, rather that just the target's parents. -n now implies -d2, so that you see what's happening. The debug level can be subsequently overridden. New -v to dump version. 4.3.2. Rules: New ALWAYS rule behaves like -t: always builds target. New EXIT rule makes it possible to raise a fatal error. New LEAVES rule which say target depends only on the update times of the leaf sources. New NOUPDATE rule says built targets only if they don't exist. NOTIME has been renamed NOTFILE, to more accurately reflect its meaning (it says a target is not to be bound to a file). 4.3.3. Variables: New special variable JAMSHELL: argv template for command execution shell. Variables, both normal and target-specific, can have their value appended with the syntax "var += value" or "var on target += value". "?=" is now synonymous with "default =". Imported enviroment variable values are now split at blanks (:'s if the variable name ends in PATH), so that they become proper list values. 4.3.4. Misc: Files to be sourced with "include" are now bound first, so $(SEARCH) and $(LOCATE) affect them. They still can't be built, though. New modifier on "actions": "existing" causes $(>) to expand only those files that currently exist. 4.3.5. Bug fixes: When scanning tokens known to be argument lists (such as the arguments to rule invocations and variable assignment), the parser now tells the scanner to ignore alphabetic keywords, as all such lists terminate with punctuation keywords (like : or ;). This way, alphabetic keywords don't need to be quoted when they appear as arguments. The scanner has been fixed to handle oversized tokens, unterminated quotes, unterminated action blocks, and tokens abutting EOF (i.e. a token with no white space before EOF). The progress report "...on xth target..." used to count all targets, rather than just those with updating actions. Since the original pronouncement of targets to be udpated included only those with updating actions, the progress report has been changed to match. 'If' conditionals now must be single arguments. Previously, they could be zero or more arguments, which didn't make much sense, and made things like 'foo == bar' true. The comparison operator is '=', and '==' just looked like the second of three arguments in the unary "non-empty argument list" conditional. Header files indirectly including themselves were mistakenly reported as being dependent on themselves. Recursing through header file dependencies is now done after determining the fate of the target. The variable expansion support was expanding $(X)$(UNDEF) as if it were $(X). It now expands to an empty list, like it should. The UNIX version of file_build() didn't handle "dir/.suffix" right. Now it does. The VMS command buffer was assumed to be as large as 1024 bytes, which isn't the case everywhere as it is related to some weird quota. It has been lowered to 256. $(>) and $(<) wouldn't expand in action blocks if the targets were marked with NOTIME. Now they expand properly. Malloc() return values are now checked. The variable expansion routine var_expand() is now a little faster, by taking a few often needed shortcuts. The VMS version of file_build() used the wrong length when re-rooting file names that already had directory compoents. This was fixed. Various tracing adjustments were made. 5. Limitations/Known Bugs The new Windows/NT support has only been marginally tested. It is dependent on certain variables being set depending on which compiler you are using. You'll need to look in the file Jambase and see what variables are expected to be set. The VMS support has been tested, courtesy of the DEC guest machine, but has not been hammered fully in release 2.0. It was used quite a bit in Jam 1.0. Jam clean when there is nothing to clean claims it is updating a target. Because the include statement works by pushing a new file in the input stream of the scanner rather than recursively invoking the parser on the new file, multiple include statements in a rule's procedure causes the files to be included in reverse order. If the include statement appears inside an if block, the parser's attempt to find the else will cause the text of the included file to appear after the first token following the statement block. This is rarely what is intended. In a rule's actions, only $(<) and $(>) refer to the bound file names: all other variable references get the unbound names. This is a pain for $(NEEDLIBS), because it means that library path can't be bound using $(SEARCH) and $(LOCATE). With the -j flag, errors from failed commands can get staggeringly mixed up. Also, because targets tend to get built in a quickest-first ordering, dependency information must be quite exact. Finally, beware of parallelizing commands that drop fixed-named files into the current directory, like yacc(1) does. A poorly set $(JAMSHELL) is likely to result in silent failure.
# | Change | User | Description | Committed | |
---|---|---|---|---|---|
#69 | 25730 | Jason Gibson | Source update for the 2.6.1 release. | ||
#68 | 16008 | Matt Attaway | Update Jam release notes for Jam 2.6 release | ||
#67 | 9954 | laurette | Update for release of 2.6. | ||
#66 | 4409 | Perforce staff |
Drop 'rc3' moniker of jam-2.5. Folded rc2 and rc3 RELNOTES into the main part of the document. Built new tar/zip balls. Only 16 months later. |
||
#65 | 3070 | Perforce staff |
Two Jambase changes: one fix, one enhancement. The fix: if a Jamrules invoked SubDir to establish other roots, and that Jamrules isn't in the current directory, the roots it established were wrong. The enhancement: Undocumented support for SUBDIRRULES, user-provided rules to invoke at the end of the SubDir rule, and SUBDIRRESET, SUBDIR variables to reset (like SUBDIRC++FLAGS, SUBDIRHDRS, etc) for each new SubDir. Bug fix and change documented only in RELNOTES. |
||
#64 | 3069 | Perforce staff |
Remove temp .bat files created on NT. Bug fix documented in RELNOTES. |
||
#63 | 3057 | rmg |
Fix mysterious rebuild problem: in an attempt to make 'jam -dc' output report headers updates more accurately, internal (header collection) targets were being bound as T_BIND_PARENTS so that they could carry the timestamp of the actual source file. But that caused the fate of the internal node to be marked as T_FATE_NEEDTMP if anything they included was newer, and that easily happens among header files (something is always newer than something else). Now internal targets carry their parents time, but with T_BIND_UNBOUND, like other NOTFILE targets. Bug fix documented in RELNOTES. === computer:1666: Change 43415 by seiwald@play-seiwald on 2003/04/04 09:36:25 |
||
#62 | 2852 | rmg |
SubDir's support for an externally set TOP variable was broken in 2.5rc1. It now works as it did in 2.4. Further, using SubDir to include a subpart of an SubDir tree now works. Previously, you could only include the root of another SubDir tree. For example, SubDir ALL src builds ; SubInclude ALL src server support ; Essentially includes the ../server/support/Jamfile, without getting confused as to the current directory. Bug fix documented in RELNOTES. === computer:1666: Change 42285 by seiwald@waffle-cyg-seiwald on 2003/03/05 08:52:10 |
||
#61 | 2851 | rmg |
Minor Cray porting: make hashitem()'s key value unsigned so we're guaranteed no integer overflows. Porting change. === computer:1666: Change 41931 by seiwald@play-seiwald on 2003/02/25 13:10:31 |
||
#60 | 2850 | rmg |
Mac classic MPW Codewarrior 7 upgrades: minor compiling issues, new paths in Jambase for libraries and includes, and separate out GenFile1 that sets PATH for UNIX only, as it does't work under MPW (or anything other than with sh). Porting change, noted in RELNOTES. === computer:1666: Change 41285 by evalynne@evalynne-bld-mac on 2003/02/05 11:47:19 |
||
#59 | 2849 | rmg |
WinXP IA64 porting: set MSVCNT to the root of the SDK and MSVCVer to Win64. Jam: change handle type to long long (too much to include windows.h?); bury IA64 in the library path in Jambase. timer.cc: use long long for timer. Also pulled over main's Jamsh.bat and extended it to handle the IA64 SDK environment when MSVCVer=Win64. Finally wrote a TarBall rule for NT that works (still uses tar and cp, but written in cmd syntax), so 'jam api' works. Porting changes, noted in RELNOTES. === computer:1666: Change 40867 by perforce@tridium on 2003/01/24 09:01:01 |
||
#58 | 2845 | rmg |
Make jam 2.5rc2 build with BorlandC 5.5 Porting Change, note in RELNOTES. === computer:1666: Change 40450 by anton@anton-diogenes on 2003/01/21 10:26:05 |
||
#57 | 2615 | rmg |
Remove NT FQuote rule, as (at least with the command shell I use), \" is required to pass quotes on the command line. Fix to bug introduced by 2.5rc1. Noted in the RELNOTES (for the moment) as a change between 2.5rc1 and 2.5rc2. === computer:1666: Change 40127 by seiwald@tricks-seiwald on 2003/01/14 16:20:25 |
||
#56 | 2614 | rmg |
Fix to includes of includes not being considered, broken by 2499. Details taken from email to jamming: The challenge is to make included includes appear as direct includes, so that they get considered. This used to work by recursion, because each TARGET node computed the summary of its includes -- the hfate and htime -- along with the summary of its dependents -- the fate and time. Alas, that previous arrangement confused make1() into treating headers as direct dependencies during the build phase. A.o | A.c -- A.h (Read depends down, includes across) (Failed build of A.h aborts build of A.c) The fix to the confused make1() problem was to consolidate the special handling of includes by having make0() tack onto a target's list of dependendies any of the target's dependents' includes. Unfortunately, this fix did not recurse: if the target's dependents' includes included other files, those files were not added to the target's dependencies. A.o | A.c -- A.h -- B.h -- C.h is rewritten to: A.o | \ A.c A.h -- B.h -- C.h (A.o depends on A.h, but not B.h or C.h. This is the current, broken state of jam 2.5rc1.) Matt's bugfix added some recursion at this point, by transitively appending includes' includes onto the includes chain. But, as he found out (and I did before), this can slow make0() down considerably, as typically header files all include each other and you wind up with lots of really long chains. A.o | A.c -- A.h -- B.h -- C.h is rewritten to: A.o | A.c -- A.h -- B.h -- C.h - B.h - C.h - C.h (Matt's fix: if the .h files include each other, the includes chains get very long.) The final(?) fix I have is relatively simple, but is an extra step: to have make0() replace a target's includes chain with a single pseudo-target whose dependencies are the original target's includes. That pseudo-target gets passed to make0(), which then recursively consolidates its fate and time. This then makes a target's includes fate and time available in a single target hanging off the original target. A.o | A.c -- A.h -- B.h -- C.h is rewritten to: A.o | \ A.c A.c-includes | \ A.h A.h-includes | \ B.h B.h-includes | C.h (New pseudo-target xxx-includes recursively consolidates fate and time of all included targets.) While this new scheme does add a node for every include file, it is linear, rather than exponential, and the time is pretty much neglible. User-visible bugfix not documented, because there is no place in RELNOTES for release-candidate fixes. Bumped patchlevel to 2.5rc2. |
||
#55 | 2613 | rmg |
Fix 'actions updated' broken by 2487. The idea was that dependents of NOTFILE targets (like "all") shouldn't be T_FATE_NEWER (newer than their parents), but instead left as T_FATE_STABLE. That way the new 'jam -dc' (display "causes") option didn't report spurious "newer" targets. NOTFILE targets have a 0 timestamp, making their children always look newer, and the change was to suppress this. But instead of checking for NOTFILE parents, the code checked for missing parents. And that broke 'actions updated', because it relied on T_FATE_NEWER even if the parent was missing. Now it checks for p->binding != T_BIND_UNBOUND (the mark of a NOTFILE target), rather than p->binding == T_BIND_EXISTS (the mark of an existing real target). === computer:1666: Change 39756 by seiwald@play-seiwald on 2003/01/03 14:53:11 |
||
#54 | 2612 | rmg |
Fix three uninitialized memory reads: 1. var_expand() wasn't checking its input length when testing for the "$(<)" shortcut. 2. Solaris does a strlen() of its sscanf() input, and ar_hdr wasn't terminated. 3. Solaris archive entries / and //xxx (the string table) generated calls to time_enter() with an uninitialized file name. Bug fixes documented in RELNOTES. === computer:1666: Change 39713 by seiwald@play-seiwald on 2003/01/02 14:46:07 |
||
#53 | 2568 | rmg | Update the release date in the RELNOTES. | ||
#52 | 2565 | rmg |
Put . in PATH for GenFile so jam can build itself without . in the global PATH. Porting change documented in RELNOTES. === computer:1666: Change 39602 by seiwald@play-seiwald on 2002/12/27 18:10:16 |
||
#51 | 2564 | rmg |
Use MSVCDIR instead of MSVCNT if MSVCNT isn't set. Microsoft changed the name to MSVCDIR in VC 6.0. Porting change documented in RELNOTES. === computer:1666: Change 39600 by seiwald@play-seiwald on 2002/12/27 18:06:42 |
||
#50 | 2563 | rmg |
Big AIX archive support. Untested. Bug fix documented in RELNOTES. === computer:1666: Change 39590 by seiwald@play-seiwald on 2002/12/27 16:40:07 |
||
#49 | 2562 | rmg |
Grist jam temp file names with pid on NT so as to ensure uniqueness. Bug fix documented in RELNOTES. === computer:1666: Change 39577 by seiwald@tricks-seiwald on 2002/12/27 15:55:21 |
||
#48 | 2561 | rmg |
RMDIR definition for NT and a default to $(RM) everywhere else. New feature documented in RELNOTES. === computer:1666: Change 39570 by seiwald@play-seiwald on 2002/12/27 15:32:38 |
||
#47 | 2560 | rmg |
Suppress last blank in variable expansion in actions. I really don't think this could possibly hurt anything. Bug fix documented in RELNOTES. === computer:1666: Change 39567 by seiwald@play-seiwald on 2002/12/27 15:23:10 |
||
#46 | 2559 | rmg |
Fix 'var on target ?= value' so that var is only set if it did not have a target-specific value. Previously, it would just overwrite the var's value. Bug fix documented in RELNOTES. === computer:1666: Change 39566 by seiwald@play-seiwald on 2002/12/27 14:44:01 |
||
#45 | 2557 | rmg |
Shuffle mechanism for optional Jamrules includes: now no error message is issued for a missing include file marked with NOCARE. Previously, we used Glob to try to find the optional Jamrules files, but on VMS that doesn't work so well: Glob returns all uppercase file names with .'s at the end, which doesn't match "Jamrules" at all. The NOCARE part is a user-visible change documented in RELNOTES. === computer:1666: Change 39273 by seiwald@waffle-cyg-seiwald on 2002/12/19 22:44:03 |
||
#44 | 2529 | rmg |
Fix "on target" variables during header scan, from Matt Armstrong. Setting target-specific variables while under the influence of the target's target-specific variables caused the _global_ values to be modified. This happened both during header file scanning and with the "on target statement" syntax. The manifestation of this was if a file #included itself, HdrRule would accidentally set HDRRULE/HDRSCAN globally, and then all files (executables, etc) would get scanned for includes. While this borrows from Matt's fix, it is a slightly different implementation. User visible fix documented in RELNOTES. === computer:1666: Change 39095 by seiwald@play-seiwald on 2002/12/17 14:00:58 |
||
#43 | 2513 | rmg |
Set the version number and patchlevel for the 2.5rc1 release. In RELNOTES, don't document the different behavior when a "return" is not the last statement executed in rule definition as a non-backward-compatible change (in the "Compatibility" section). |
||
#42 | 2509 | rmg | Fix an old typo. | ||
#41 | 2508 | rmg | First pass edit on the RELNOTES for 2.5. | ||
#40 | 2503 | rmg |
No longer link jam with advapi32.lib -- it isn't needed and it keeps moving. No functional change. === computer:1666: Change 38674 by seiwald@tricks-seiwald on 2002/12/10 12:02:55 |
||
#39 | 2502 | rmg |
Update RELNOTES to reflect the addition of history comments, the removal of the /MR name, and the rewrite of Jam.html. === computer:1666: Change 38641 by seiwald@play-seiwald on 2002/12/09 22:04:15 |
||
#38 | 2499 | rmg |
Fix 'includes' support so that included files aren't treated as direct dependencies during the command execution phase. If an included file failed to build, make1() would bypass the including file. Now make0() appends each child's 'includes' onto its own 'depends' list, eliminating 'includes'-specific code in make0() and make1(). This not only fixes the bug, but removes some complexity as well. Bug fix documented in RELNOTES. === computer:1666: Change 38399 by seiwald@play-seiwald on 2002/12/03 16:00:40 |
||
#37 | 2497 | rmg |
Jambase no longer announces the compiler on Windows. It doesn't announce anything else. User visible change documented in RELNOTES. === computer:1666: Change 38194 by seiwald@tricks-seiwald on 2002/11/25 15:02:57 |
||
#36 | 2494 | rmg |
Remove bogus search() call in 'on' statement processing, as pointed out by Ingo on the jamming mailing list. === computer:1666: Change 38000 by seiwald@play-seiwald on 2002/11/18 12:19:24 |
||
#35 | 2492 | rmg |
Make jam on solaris sparc set OSPLAT to sparc. This will change the build directory for solaris executables to bin.solarissparc! User visible change documented in RELNOTES. === computer:1666: Change 37638 by seiwald@play-seiwald on 2002/11/05 14:48:06 |
||
#34 | 2491 | rmg |
Some consting in jam to make it more compilable by C++ compilers. No functional change. === computer:1666: Change 37433 by perforce@perforce on 2002/10/30 16:08:51 Recreational const-ing of jam, for compilers that don't allow "string" to be passed as a non-const char *. This included a few places where we were modifying what could possibly have been read-only storage, oddly enough. No functional change. === computer:1666: Change 37602 by seiwald@play-seiwald on 2002/11/04 17:25:40 |
||
#33 | 2490 | rmg |
Jam langauge work: make 'return' actually return from the rule, rather than just setting the return value. Introduce new break/continue statements for managing loops. User visible change to be documented in Jam.html. === computer:1666: Change 37200 by seiwald@play-seiwald on 2002/10/22 15:41:28 Gross rework of Jam.html documentation, including: - the description of parameters for rules - description of -g flag - a new description of targets - more about rules and their return values - better separation of rules and updating actions - putting borders around the tables (Undocumented) change to documentation. === computer:1666: Change 37551 by seiwald@waffle-cyg-seiwald on 2002/11/03 23:17:12 Document jam's new and working break/continue/return statements. === computer:1666: Change 37574 by seiwald@play-seiwald on 2002/11/04 13:13:01 |
||
#32 | 2488 | rmg |
Remove the /MR suffix from Jam. === computer:1666: Change 37146 by seiwald@play-seiwald on 2002/10/21 15:23:18 |
||
#31 | 2487 | rmg |
Subtle jam adjustments in make0() to improve tracing output, and some new tracing output (-dc: show 'causes'). First, make0() now takes a pointer p to the parent, rather than just the parent time ptime. This helps when setting the T_FATE_NEWER: it's only set if the target is newer than the parent _and the parent exists_. Otherwise, T_FATE_NEWER gets set for all dependenents of NOTFILE targets. When determining fate, don't make the final check to see if we need to downgrade T_FATE_NEWER to T_FATE_STABLE: fate is always T_FATE_NEWER or T_FATE_STABLE, so we might as well just set it to T_FATE_STABLE anyhow. New -dc output shows what targets are causing things to get rebuilt. User visible change documented in RELNOTES. === computer:1666: Change 36605 by seiwald@play-seiwald on 2002/09/28 18:26:32 |
||
#30 | 2486 | rmg |
Fooling around with jam's -d flag, to make it possible to specify useful display output without turning on loads of debugging crud. New -dd flag to display dependencies. Provisional changes not yet documented in jam.html. === computer:1666: Change 36374 by seiwald@play-seiwald on 2002/09/19 15:17:20 Jam -d change: the message "...using xxx..." now only shows up with -da, rather than in the default output. It made it hard to see what was happening when there were a lot of temp files lying around. User visible change documented in RELNOTES. === computer:1666: Change 36430 by seiwald@play-seiwald on 2002/09/23 11:34:12 Put jam -dx flags into 'jam -h'. Change to undocumented behavior (jam -h's output). === computer:1666: Change 36551 by seiwald@play-seiwald on 2002/09/26 14:39:54 Document jam's new -d debug flags. === computer:1666: Change 37367 by seiwald@waffle-cyg-seiwald on 2002/10/28 16:03:46 jam -n now implies -dax, just as the old jam -n implied -d2. Change to unreleased functionality. === computer:1666: Change 37550 by seiwald@waffle-cyg-seiwald on 2002/11/03 23:12:15 |
||
#29 | 2485 | rmg |
Beos updates for jam from "Ingo Weinhold" <bonefish@cs.tu-berlin.de>. Porting change. === computer:1666: Change 35634 by seiwald@play-seiwald on 2002/08/16 11:42:39 |
||
#28 | 2484 | rmg |
Rework make0() to consider headers when deciding whether to update a temporary source file. This makes it possible to have generated, temporary source files that must be regened because an included header has been updated. === computer:1666: Change 35018 by seiwald@play-seiwald on 2002/07/17 10:41:35 Followon to 35018: actually update a "needtmp" target (duh). === computer:1666: Change 35023 by seiwald@play-seiwald on 2002/07/17 11:10:16 Fix to 35018, which handled out-of-date header files built from temporary sources. Unfortunately, the change made circular header dependencies (which are legion) get reported. Now the circular dependency reporting is done in the call to make0() rather than at the beginning of make0(), and step 3b (the header recursion) simply skips that check. Change to unreleased behavior. === computer:1666: Change 36247 by seiwald@waffle-cyg-seiwald on 2002/09/16 16:15:10 |
||
#27 | 2483 | rmg |
Jam macosx changes: use 'ar' instead of libtool, as ar can do incremental updates. New support for the BSD style long filenames in archives. Porting change. === computer:1666: Change 34993 by seiwald@oswald-seiwald on 2002/07/16 12:01:32 Fix archive scan (solaris and FreeBSD style) broken by change #34993. I _think_ all three forms of archive entries now work again (traditional, solaris string table, and macosx #1/xxxx ). === computer:1666: Change 35027 by seiwald@play-seiwald on 2002/07/17 11:31:20 Fix archive scan supposedly fixed in 35027 (which was to fix 34993). Now it works on MacOSX again. === computer:1666: Change 35057 by perforce@cardtrick on 2002/07/17 16:11:01 |
||
#26 | 2481 | rmg |
Put . in the path of jam0, at long last. === computer:1666: Change 34039 by seiwald@tricks on 2002/06/03 11:05:04 Jam relnote about $(EXENAME) in the Makefile. User visible change being documented in RELNOTES. === computer:1666: Change 34517 by seiwald@play-seiwald on 2002/06/22 00:19:32 |
||
#25 | 2480 | rmg |
Rewrite jam's SubDir rule to allow multiple roots. Infrastructure change. === computer:1666: Change 33320 by seiwald@thin on 2002/05/13 10:10:50 |
||
#24 | 1658 | rmg | 2.4rc2 becomes 2.4 (final). | ||
#23 | 1617 | rmg | Branding 2.4rc2 | ||
#22 | 1580 | rmg |
Changes for 2.4beta. Yes, Christopher, I brashly said "2.4-dev WAS the beta" at tea yesterday, but I've since come to my better instincts (got chicken?), so this one will say rmg $ jam -v Jam 2.4beta. OS=LINUX. Copyright 1993-2002 Christopher Seiwald. for a week or two, at least. |
||
#21 | 1572 | Perforce staff |
Fix InstallInto to handle multiple sources properly, thanks to Stephen Goodson. Bug fix documented in RELNOTES. |
||
#20 | 1571 | Perforce staff |
MINGW port by Max Blagai. Porting change documented in RELNOTES. |
||
#19 | 1570 | Perforce staff |
New SoftLink rule, courtesy of David Lindes. User visible change documented in RELNOTES. |
||
#18 | 1566 | rmg |
Correct Michael Allard's name in the RELNOTES. doc change only. |
||
#17 | 1565 | Perforce staff |
Suffix yyacc generated symbols with _t to ensure they don't conflict with other symbols, as RULE was. |
||
#16 | 1553 | Perforce staff |
New -g flag to reorder targets so that they are built according to their sources update times, rather than simply the order in which they appear in the Jamfiles. This flag is experimental: it isn't clear if sorting on target time is adequate to ensure newest sources are seen first, and it isn't clear that this is actually useful at all. Undocumented except for a note in RELNOTES. |
||
#15 | 1545 | Perforce staff |
Handle TEMPORARY targets with multiple parents better (but not perfectly). Previously, a TEMPORARY target that had no timestamp of its own (t->time == 0, due to the target being missing) took on its parent's timestamp ptime. The target's timestamp is used for four things: 1. It is passed as the target's timestamp to its children, in case they are also TEMPORARY targets (and thus need their parent's timestamp). 2. It is compared against the target's children's timestamps, to see if the target is out of date. 3. It is compared against the target's parent time, to see if the target is newer than its parent (for 'actions updated'). 4. It is saved as the target's timestamp, for when the target's parents need to see if they are out of date. The problem was with (4) and multiple parents: the first parent's timestamp would become the target's, which could then inadvertantly outdate another parent older than the first but otherwise up-to-date. With this change a missing TEMPORARY target is left with a zero timestamp, and the four above cases are modified for T_BIND_PARENT targets as such: 1. The parent's timestamp is passed directly to the target's children. 2. The parent's timestamp is compared directly against the target's children's timestamps. 3. The parent's timestamp is no longer compared against the target's own, as it could never be older. 4. The target's timestamp remains zero (or the newest of the target's children's), so as not to outdate other parents. Oddly enough, this change mostly just removes the setting of 't->time = ptime' (leaving t->time zero), and adds only a little extra logic to use ptime directly when needed. It also removes the 'pbinding' parameter to make0(), as its only use (to indicate if the parent was missing) can better be divined from !ptime. Perfection, as they say, isn't achieved when there is nothing more to add, but nothing more to take away. We'll see if I removed enough this time. Bug fix documented in RELNOTES. |
||
#14 | 1535 | Perforce staff | Define YYMAXDEPTH to 10000 to handle right-recursive rules. | ||
#13 | 1533 | Perforce staff | Relnotes for Match, indirect rules, and 'actions together' fix. | ||
#12 | 1374 | rmg |
Fixed broken $(v[1-]), which always returned an empty expansion. Thanks to Ian Godin <ian@sgrail.com>. This change is Ian's original, "surgical-strike" fix. We may yet replace it with something different, but this at least gets the mainline fixed for now. |
||
#11 | 1356 | rmg |
Tweak about Glob description, as suggested by n.giger@netstal.com |
||
#10 | 1346 | rmg |
Add an option that gets Jam to exit as soon as any target fails (as if it had received an "interrupt") Integrates Change 233 by Peter Glasscock. Added to Jam.html & RELNOTES - rmg |
||
#9 | 1344 | rmg |
Under NT, default $(MV) to "move /y" in Jambase. Logically (and in the integ hhistory) integrates Change 390 by Eric Scouten. Cc: eric@scouten.com |
||
#8 | 1319 | rmg |
Jam 2.3 + Perforce's internal changes. This change is a drop of the Perforce internal Jam changes since the 2.3 public release. The individual changes represented herein are preserved in the //guest/richard_geiger/intjam/ branch. The intent of this drop is to provide a base, from which other contributors' Jam branches may be integrated into. It is not intended to become a packaged release in this state. We will be integrating changes from other users prior to creating the next packaged release. Please refer to the src/RELNOTES file for an overview of the changes present in this integration. - Richard Geiger Open Source Engineer at Perforce |
||
#7 | 556 | Perforce staff |
Patch release jam 2.3.2: reverse NOCARE change. NOCARE once again applies to targets with source and/or actions, so that buildable header files get built. |
||
#6 | 486 | Perforce staff |
Jam 2.3. See RELNOTES for a list of changes from 2.2.x. Just about every source file was touched when jam got ANSI-fied. |
||
#5 | 211 | Perforce staff | Jeff Sickel's Mac OS X port. | ||
#4 | 5 | Perforce maintenance | Jam/MR 2.2.4 (HDRPATTERN, JAMUNAME, JAMSHELL, plus misc tweaks) | ||
#3 | 4 | Perforce maintenance | Jam/MR 2.2.2 (AmigaOS support) | ||
#2 | 3 | Perforce maintenance | Jam/MR 2.2.1 (fix for NT handle leak) | ||
#1 | 2 | laura | Add Jam/MR 2.2 source |