humungus - gerc

Added tag v0.1.4 for changeset 8802a38593a7

bump gozstd

switch to zstd that supports !cgo

fix colorwriter to not hold buffer it doesn't own

m

speed up diff by using the previous revision we already have.

try coloring log diffs

one less pointer indirection

clean up a bit

spoiler: it's actually slower

make revlog "faster" by copying less data

compute more deeper sha1

compute sha1 for changes

dumb global snuck in

note compression (

partial implementation of status

Added tag v0.1.3 for changeset 08957fecb2df

zstd support

need to read tag file from within store, not filesystem

add license

add the colorizer code too

m

colorized diff

basic usage

typo

Added tag v0.1.2 for changeset a6f9e572fd86

if delta data is empty, no deltas

align tags output

don't count tags that have been deleted

cache frequently used fds

Added tag v0.1.1 for changeset 5a4e4fb32602

need to handle the .d datafile case now

print a hint with the diff

slightly easier to follow

Added tag v0.1.0 for changeset dcd9019e67c3

explain why we need another copy

print parent revisions when not obvious

use per file last change timestamp

forgot about usage

add diff command, for comparing work files vs repo

ancestors go the other way

a go.mod file

incomplete and wrong ancestor adjustment for revlog

move all argv string parsing into frontend.
programmatic args for library functions

a hack to walk backwards to find last existing rev for annotate

add a package comment for main

split big file into small files

add annotate support

rename a variable to be clearer

add go doc for public functions

-r last() is also easy to handle

add tags command

handle a few special cases for filenames

replace writestring with fprintf

tune up the data/metadata handling a bit

works a little better if the caller adds the extra newline

refine change printing code

print tags in log

parse tags file and be more flexible about hash lookups

remove bad check for empty files

remove metadata from revlog data. ignored for now.

simpler and safer to make a function to split bytes to lines

check len of right slice

an empty file isn't missing its newline

not allowed to dig into .hg either

check that filenames are inside the repo

reverse logic of missingnl to be hasnl

it's a feature not a bug

can binary search the manifest instead of scanning

try finding unqiue lines, but still do the first pass with anything we find

clear revlog cache after close

parse limit(range, limit) revnums too

start parsing revnums in date format. today and yesterday.
this is starting to get troublesome.

leave the message as an array

formatting tweaks

cache changes after parsing, but be careful not to cache diffs

need to slice off unused end of data after applying deltas

nevermind, the ulen is uncompressed full data len, not just this chunk

using a scratch alice between delta application massively reduces allocs

allocs profile is more interesting

can presize uncompress buffer

print the correct dates in changesets

limiting to one cpu is much faster

fix calculation of offsets in unidiff

another peephole fixup: sometimes chunks get combined + then -. reverse them.

revert boring check, doesn't help and makes some diffs worse

missing new line in printf

skip boring lines for matches

allow cat of any revision

fix missing context line

test for trailing context

understand tip to mean -1

rename author to user

a few cleanups

map linkrevs to index

slowly working diffing prev revisions into shape

fold diff fetching into main change loop

fix check for corrupted delta

use a bufio for reading files

give parsedeltas the zero copy treatment too

perf improvement: much less copying when applying deltas

perf improvement: cache all revlogs instead of reopening

global flags for profiling

add a few tests for cases that have tricked patience in the past

further fixes and refinements to diff

improved diff creation

there can be more than one boring line, don't skip too far ahead

a few tweaks to the peephole for prettier diffs

debug print keeps sneaking in

only uncount context lines in one loop

eat -u for diff compat

some fixes for the amount of context in diffs

missing newline in error message

use diff now that it works instead of toying with deltas

there's no diff if there's no diff

try recursing one time

update comment

the one line skipper causes more trouble than its worth

paste error in bounds check

not quite correct handling of files missing final newlines

struct of slices is too much trouble to work with.
back to slice of structs

ok, getting silly, but add a peephole optimizer for a few more blank line cases

handle a few special cases when creating diffs

possible for a p2 line to be at 0

actually save space by using byte instead of rune

use two slices instead of wasting padding

add an implementation of patience diffing algorithm

fix revision fetching with a filename

negative revisions are pretty easy

parse hex revs too

start parsing revnums as ranges

ah ha! baserev chain can be linear or not linear depending on flag.

don't go past end getting forw context

storename: _ is __

fixes for context diffs

baserev is linear? real bug was skipping one?

fix a bug where we would skip to base for short chains

want the list of files in the changeset, not the manifest

cheating method can't compute diffs from full revisions

add the spec for known file formats

fix storename escaping

the baserev chain may not be linear

copyright and readme

allow closing a repo

sometimes the baserev isn't the baserev...

support name args for files

smooth some delta operations

make a repo object instead of global

quick conversion to package and main

good enough revision control.
good enough to run a few commands anyway.