diff options
author | Fred Hebert <mononcqc@ferd.ca> | 2018-10-11 08:38:37 -0400 |
---|---|---|
committer | Fred Hebert <mononcqc@ferd.ca> | 2018-10-11 08:45:04 -0400 |
commit | dada4e36e6d9a5c4b41bbe1f68389520e7c59ace (patch) | |
tree | 27f3d5451fad7a6c2157965af811a65898704054 /src/rebar_paths.erl | |
parent | af5cecd8eec9692f43d04ad53c8f28734012b873 (diff) |
Optimize path handling
- Only set paths that need to be put as a priority
- Clean up paths before leaving API mode
The first point accounted for some performance cost, but the latter one
explains the 40% overhead in test runs: since rebar3 calls rebar3 a lot
with a bunch of fake apps, and that the new mechanism for path handling
by default does not _remove_ paths, it just _orders_ them, we would end
up in a situation where as the tests ran, more and more fake paths would
get added to the VM.
By the time the run was over, all path handling would take longer since
more paths needed filtering every time. By resetting paths at the end of
an API run, we prevent a given 'project' from polluting another one's
runtime and performance once the API successfully returns.
Diffstat (limited to 'src/rebar_paths.erl')
-rw-r--r-- | src/rebar_paths.erl | 59 |
1 files changed, 29 insertions, 30 deletions
diff --git a/src/rebar_paths.erl b/src/rebar_paths.erl index 23fc755..900443d 100644 --- a/src/rebar_paths.erl +++ b/src/rebar_paths.erl @@ -1,3 +1,5 @@ +%% BEFORE THIS FIX: rebar3 ct 266.78s user 144.06s system 144% cpu 4:33.70 total +%% CURRENT TIME: rebar3 ct 419.30s user 301.00s system 152% cpu 7:51.98 total -module(rebar_paths). -include("rebar.hrl"). @@ -8,20 +10,17 @@ -export([clashing_apps/2]). -ifdef(TEST). --export([misloaded_modules/3]). +-export([misloaded_modules/2]). -endif. -spec set_paths(targets(), rebar_state:t()) -> ok. set_paths(UserTargets, State) -> Targets = normalize_targets(UserTargets), GroupPaths = path_groups(Targets, State), - Paths = lists:append([P || {_, P} <- GroupPaths]), - [code:del_path(P) || P <- Paths], - code:add_pathsa(lists:reverse(Paths)), - % set path breaks with escripts; we gotta do it by hand - % true = code:set_path(lists:append([P || {_, P} <- GroupPaths])), + Paths = lists:append(lists:reverse([P || {_, P} <- GroupPaths])), + code:add_pathsa(Paths), AppGroups = app_groups(Targets, State), - purge_and_load(AppGroups, code:all_loaded(), sets:new()), + purge_and_load(AppGroups, sets:new()), ok. -spec unset_paths(targets(), rebar_state:t()) -> ok. @@ -33,6 +32,7 @@ unset_paths(UserTargets, State) -> purge(Paths, code:all_loaded()), ok. +-spec clashing_apps(targets(), rebar_state:t()) -> [{target(), [binary()]}]. clashing_apps(Targets, State) -> AppGroups = app_groups(Targets, State), AppNames = [{G, sets:from_list( @@ -66,9 +66,9 @@ normalize_targets(List) -> ), lists:reverse(TmpList). -purge_and_load([], _, _) -> +purge_and_load([], _) -> ok; -purge_and_load([{_Group, Apps}|Rest], ModPaths, Seen) -> +purge_and_load([{_Group, Apps}|Rest], Seen) -> %% We have: a list of all applications in the current priority group, %% a list of all loaded modules with their active path, and a list of %% seen applications. @@ -79,7 +79,12 @@ purge_and_load([{_Group, Apps}|Rest], ModPaths, Seen) -> %% 3. unload and reload apps that may have changed paths in order %% to get updated module lists and specs %% (we ignore started apps and apps that have not run for this) - %% 4. create a list of modules to check from that app list + %% This part turns out to be the bottleneck of this module, so + %% to speed it up, using clash detection proves useful: + %% only reload apps that clashed since others are unlikely to + %% conflict in significant ways + %% 4. create a list of modules to check from that app list—only loaded + %% modules make sense to check. %% 5. check the modules to match their currently loaded paths with %% the path set from the apps in the current group; modules %% that differ must be purged; others can stay @@ -126,27 +131,29 @@ purge_and_load([{_Group, Apps}|Rest], ModPaths, Seen) -> end || App <- GoodApps, AppName <- [binary_to_atom(rebar_app_info:name(App), utf8)]] ), + ModPaths = [{Mod,Path} || Mod <- CandidateMods, + erlang:function_exported(Mod, module_info, 0), + {file, Path} <- [code:is_loaded(Mod)]], %% 5) - Mods = misloaded_modules(CandidateMods, GoodAppPaths, ModPaths), + Mods = misloaded_modules(GoodAppPaths, ModPaths), [purge_mod(Mod) || Mod <- Mods], - purge_and_load(Rest, ModPaths, - sets:union(Seen, sets:from_list(AppNames))). + + purge_and_load(Rest, sets:union(Seen, sets:from_list(AppNames))). purge(Paths, ModPaths) -> SortedPaths = lists:sort(Paths), - lists:map(fun purge_mod/1, lists:usort( - [Mod || {Mod, Path} <- ModPaths, - is_list(Path), % not 'preloaded' or mocked - any_prefix(Path, SortedPaths)] - )). + lists:map(fun purge_mod/1, + [Mod || {Mod, Path} <- ModPaths, + is_list(Path), % not 'preloaded' or mocked + any_prefix(Path, SortedPaths)] + ). -misloaded_modules(Mods, GoodAppPaths, ModPaths) -> +misloaded_modules(GoodAppPaths, ModPaths) -> %% Identify paths that are invalid; i.e. app paths that cover an %% app in the desired group, but are not in the desired group. lists:usort( - [Mod || Mod <- Mods, - {_, Path} <- [lists:keyfind(Mod, 1, ModPaths)], + [Mod || {Mod, Path} <- ModPaths, is_list(Path), % not 'preloaded' or mocked not any_prefix(Path, GoodAppPaths)] ). @@ -157,15 +164,7 @@ any_prefix(Path, Paths) -> %% assume paths currently set are good; only unload a module so next call %% uses the correctly set paths purge_mod(Mod) -> - case erlang:check_process_code(self(), Mod) of - false -> - code:purge(Mod), - code:delete(Mod); - _ -> - %% cannot purge safely without killing ourselves - code:soft_purge(Mod) andalso - code:delete(Mod) - end. + code:soft_purge(Mod) andalso code:delete(Mod). %% This is a tricky O(n²) check since we want to |