We assigned user capability characters using only lowercase ASCII letters at first, so those are the most important within Fossil: they control the functions most core to Fossil’s operation. Once we used up most of the lowercase letters, we started using uppercase, and then during the development of the forum feature we assigned most of the decimal numerals. All of the lowercase ASCII letters are now assigned. Eventually, we might have to start using ASCII punctuation and symbols. We expect to run out of reasons to define new caps before we’re forced to switch to Unicode, though the possibilities for mnemonic assignments with emoji are intriguing. 😉
The existing caps are usually mnemonic, especially among the earliest and therefore most central assignments, made when we still had lots of letters to choose from. There is still hope for good future mnemonic assignments among the uppercase letters, which are mostly still unused.
Some may question the use of ASCII character strings for capability sets instead of bitfields, which are more efficient, both in terms of storage and processing time.
Fossil handles these character strings in one of two ways. For most HTTP
hits, Fossil expands the string into a
struct full of
flags so that later code can just do simple Boolean tests. In a
minority of cases, where Fossil only needs to check for the presence of
a single flag, it just does a
strchr() call on the string
Both methods are slower than bit testing in a bitfield, but keep the execution context in mind: at the front end of an HTTP request handler, where the nanosecond differences in such implementation details are completely swamped by the millisecond scale ping time of that repo’s network connection, followed by the required I/O to satisfy the request. Either method is plenty fast in that context.
In exchange for this immeasurable cost per hit, we get human-readable capability sets.
Fossil is more trusting about the content it receives from a remote clone during sync than you might expect. Common manifestations of this design choice are:
A user may be able to impersonate other users. This can be accidental as well as purposeful.
If your local system clock is out-of-sync with absolute time, artifacts committed to that repo will appear with the “wrong” time when sync’d. If the time sync error is big enough, it can make check-ins appear to go back in time and other bad effects.
You can purposely overwrite good timestamps with bad ones and push those changes up to the remote with no interference, even though Fossil tries to make that a Setup-only operation.
All of this falls out of two of Fossil’s design choices: sync is all-or-nothing, and the Fossil hash tree is immutable. Fossil would have to violate one or both of these principles to filter such problems out of incoming syncs.
We have considered auto-shunning “bad” content on sync, but this is difficult due to the design of the sync protocol. This is not an impossible set of circumstances, but implementing a robust filter on this input path would be roughly as difficult as writing a basic inter-frame video codec: do-able, but still a lot of work. Patches to do this will be thoughtfully considered.
We can’t simply change content as it arrives. Such manipulations would change the artifact manifests, which would change the hashes, which would require rewriting all parts of the block chain from that point out to the tips of those branches. The local Fossil repo must then go through the same process as the remote one on subsequent syncs in order to build up a sync sequence that the remote can understand. Even if you’re willing to accept all of that, this would break all references to the old artifact IDs in forum posts, wiki articles, check-in comments, tickets, etc.