Outgoing InterTrac links _from_ a CVSTrac install can be accomplished with custom markup, as in:

{quote:%x}

which would be referenced via something like:

{quote:{<markup>: ticket:451 #451}} {quote:{trac: wiki:InterTrac InterTrac}}

This means we need to allow generic RSS substitutions in the header. To that end, we add %R for the URL and %F for the title (a distinct feed needs a distinct title).

Figuring out whether a page has its own RSS feed is done by looking for an "RSS" entry in the action bar. The title is built from the project name and the current page title.

Then, to override the default (timeline) RSS feed on any given page, it's just a matter of doing something like:

common_add_action_item(mprintf("tkthistory.rss?tn=%d",tn), "RSS");

and the header output logic will just pick it up. (check-in: c72535f17c user: cpb tags: trunk)

In order to allow it to work in both normal web browsing and the encoding needed for RSS (where we _need_ to encode even entities), I modified htmlize (%h) to take an "allow entities" flag, and added a new %R formatting option which disables this flag. %R is used by the RSS formatter to do the final content generation. This means anywhere wiki content is allowed someone can write é and, apparently, it works.

Keep in mind that I don't normally use anything except ASCII in my day-to-day environment, so someone who cares about this stuff should maybe test it out a bit more than myself. (check-in: 14e235cdd5 user: cpb tags: trunk)

Only one left in the table stuff in /cvstrac/view.c (check-in: f18b4a5e70 user: cpb tags: trunk)

Basically, when a user starts editting a ticket (or appending a remark), the last "changetime" is embedded in the form. At submit time, CVSTrac checks to ensure that changetime is still good and refuses to proceed. The user still has all the change visible and can copy them elsewhere, but can't "finish" the change. A "force" checkbox might be useful here?

By default, this is disabled. (check-in: 6e09701d50 user: cpb tags: trunk)

tags inside forms * changing hspace into margin * changing wrap="type" into class wraptype (this may be problematic) * adding DocType ... still we got Content-Type missing and warning with lists +

at Wiki

*: entity tag is a =quoted-string=, and hence needs quotes *: Vary:* SHOULD be included when dealing with server-negotiated responses (check-in: 74510a0008 user: cpb tags: trunk)

*: This should catch both "NULL" descriptions and zero-length strings. *: Additionally, tighten up the setup page to always remove the old description before changing anything and check the new description before setting anything. (check-in: 163a463930 user: cpb tags: trunk)

Break the two separate conditions into separate <form> areas and loosen up the condition which create the setup_user link. (check-in: d25d699321 user: cpb tags: trunk)

CVSTrac doesn't really have a way to track proper merges (which basically implies multiple CHNG.prevvers entries) but we _can_ reference merge source check-ins in the commit message. It's still not as easy as I'd like to follow GIT development, but this makes it clearer what things are merges and what commits are being pulled into the merge.

*: we need to ingest the whole commit tree as we check refs, and we need to do it such that roots get the lowest CN numbers and are databased before anything which relies on them. This means making =git_ingest_commit_tree()= truly recursive. *: drop the temporary =ci= table, since we now need to do it all in one pass *: add a function to convert multiple parents into a merge list and update the check-in comment with something like "Merged [n1], [n2], ..." *: in order to _display_ merges, we need to use =git-diff= rather than =git-diff-tree=. *: add =pxHistoryReconstructPrep= and =pxHistoryRescanPrep= callbacks to =g.scm= so we can "do things" to SCM-specific tables and whatnot before a history reconstruct or rescan. In the case of GIT, a reconstruct requires removing all the refs milestones. Other SCMs may have similar requirements.

This doesn't break database compatibility, but you _will_ have to do a reconstruct if you want to see merges properly. Unfortunately, a reconstruct may alter check-in numbers and break links in wiki text or manually created ticket cross-references. (check-in: 3219fa7f89 user: cpb tags: trunk)

The fix is to detect the missing FILECHNG records and ensure we create the refs milestone anyways. Once we've got the refs milestone with an proper object reference, we'll never try to update those refs again. (check-in: 7e9b7e50b0 user: cpb tags: trunk)

Basically, we drop the TAGS and HEADS tables entirely and exploit the fact that GIT doesn't make _any_ use of the CHNG.branch and CHNG.directory fields. CHNG.branch is meaningless since branches aren't static and a single commit might belong to any number of branches. CHNG.directory is also meaningless because GIT operates on the entire tree in a single commit. So for GIT milestones, CHNG.branch=refs/* and CHNG.directory=object.

This allows us to completely drop a couple tables (which may break reports) and simplifies a lot of queries and stuff. Because we've got other places to put things, the milestone message can also be shortened.

To upgrade an existing database, something like:

DROP TABLE heads; DROP TABLE tags; DELETE FROM chng WHERE milestone>0 AND message LIKE '%(%, commit %';

before any GIT history update will be fine.

As an added bonus, we're no longer looking directly at files in the GIT repository. We just use command-line tools. So we're probably a lot more resiliant to changes in repository layouts. (check-in: 89dfc6a081 user: cpb tags: trunk)

Use another temp table to help prune the list down to the bare minimum. There's room for other improvements, too. (check-in: a00e3770a4 user: cpb tags: trunk)

The obvious and 100% compatible fix would be something like this:

--- format.c 30 Sep 2006 11:01:30 -0000 1.91 +++ format.c 28 Jan 2007 21:33:13 -0000 @@ -1323,8 +1323,9 @@ || (sMarkup.lenType==11 && strncmp(sMarkup.zType,"rightreport",11)==0) || (sMarkup.lenType==10 && strncmp(sMarkup.zType,"leftreport",10)==0) ){ - char *zCaption = mprintf("%.*s", sMarkup.lenArgs, sMarkup.zArgs); + char *zCaption, *zCaptionOrig; char *zAlign = 0; + zCaptionOrig = zCaption = mprintf("%.*s", sMarkup.lenArgs, sMarkup.zArgs); if( sMarkup.lenType==11 ){ zAlign = "align=\"right\""; }else if( sMarkup.lenType==10 ){ @@ -1335,7 +1336,7 @@ embed_view( atoi(sMarkup.zKey), (sMarkup.zArgs==sMarkup.zKey) ? "" : zCaption, zAlign ); - free(zCaption); + free(zCaptionOrig); zText += sMarkup.lenTotal; i = 0; wordStart = lineStart = paraStart = 0;

But the whitespace skipping never can happen as the is_markup() call around the parsing code already skipped all leading whitespaces. Hence we fix the deallocation problem by just removing the unnecessary second whitespace skipping attempt. (check-in: 72db38160a user: rse tags: trunk)

Currently, we're using the authorizer to enforce restrictions on report statements. Since, internally, CVSTrac distinguishes between query and execute styles of commands, it makes some sense to add similar authorizer facilities to some of these functions as well.

This patch introduces the concept of an authorizer stack to which we can push and pop from (otherwise, interactions between query and the existing report authorizer are impossible to manage) and uses it in all the query-style functions. As a side-effect we also need to create a separate =db_execute_callback()= function. A define, USE_STRICT_AUTHORIZER, is created to "help" CVSTrac developers avoid using the from query/execute call.

The goals are multiple: to defang potential SQL injections, inadvertant data leaks, and (eventually) privledge escalations such as are currently blocked in reports. Eventually, it may also get used to help enforce protections should we implement embedded scripting. It also introduces a framework for preventing silly mistakes (i.e. CREATE/DROP table outside of schema and history updates, deleting rather than updating things, etc). (check-in: 41fc1c9f55 user: cpb tags: trunk)

common_.c:399: warning: the address of 'azLink' will always evaluate as 'true' common_.c:415: warning: the address of 'azAction' will always evaluate as 'true'

..for the following code lines...

[...] if( nLink && azLink ){ [...] if( nAction && azAction ){ [...]

...as we have...

static const char *azLink[50]; static const char *azAction[50];

...and ISO-C always implicitly makes the name of the array variable to be a pointer to the array and as the array is a static array it _always_ is a valid pointer and hence always "true" in boolean context.

OTOH we don't need the comparison here at all as the nLink and nAction check is really sufficient to know whether the array contain content. So, remove the warning by removing the useless parts from the boolean expression. (check-in: ad7cff7f48 user: rse tags: trunk)