win32: remove text man pages from Windows package The HTML versions are still packaged, and do not suffer from eoln problems. This solves issue531 (addition through subtraction)

Merge with mpm

win32: include zsh_completion in Windows package

win32: include contrib/vim in Windows package

win32: use relative path to install source Remove hard-coded path to the source repo to be packaged.

win32: move default install path inside program files All of the 'space in path' problems have been resolved, so we can install Mercurial the Windows standard location. The {pf} InnoSetup variable is locale aware.

win32: it is no longer necessary to bundle patch.exe Mercurial can use it's built-in mpatch library

win32: do not install bash hgmerge on Windows

win32: ignore all installers generated in Output/

convert: move commands definition to ease demandload job (issue 860)

Merge with crew

win32: favor local mercurial.ini over registry key Hopefully this is the best of both worlds. Multiple py2exe based installers can coexist on one machine, while source and lib based Mercurial users can still use a registry key in lieu of resorting to HGRCPATH.

Merge with stable

convert: fix SVN date parser dropping the final whole second digit

highlight: adapt to hgweb_mode refactoring

Merge with -stable

Fix bdiff test failures on AIX. The test fails when, in the call to calloc(), the number of elements (bn) is 0. In that case, calloc() on AIX will return NULL, while the code expects a valid heap pointer. Both results are permissible under C99, Unix98 etc. Work around by ensuring that at least 1 element is requested.

workaround for raw_input() on Windows Since change a3fe91b4f6eb, Mercurial has set_binary() on stdin, stdout, and stderr. On Windows, this had the side effect of causing raw_input() to emit trailing carriage returns on it's returned strings. This was causing web authentication problems.

Merge with crew-stable

utils_win32: calling SHGetPathFromIDList under NT is useless and broken The call was done under NT when running as a service (CGI script for instance) and always crash, even in interactive processes.

copy: handle rename internally - add rename parameter - don't collect copied list - report "moving" rather than "copying" - remove renamed files - don't return copied list - simplify callers

copy: move rename logic

copy: simplify inner copy - save dirstate lookup and exists check - do all fs work inside a single dryrun clause - move unlinking into inner copy section - move target dir creation - eliminate undelete mess - fix a bug on mv a -> b -> a after merge

copy: minor cleanups - add after and dryrun variables - add some comments

copy: propagate errors properly

copy: refactor okaytocopy into walkpat - rename core copy function to copyfile - move origsrc details into copyfile - turn okaytocopy loop into walkpat

copy: fix copying back with -A (issue836)

hgwebdir: refactor inner loop

hgwebdir: split out templater creation

hgwebdir: split out makeindex function, facilitate test failure diagnosis