The compiling of OTP applications is done by first topographically
sorting them according to their dependencies, deps-first. This allows
all compilation to take place in order. In the current code, the same
logic extends to top-level applications in an umbrella project.
Unfortunately, there are cases where this is not going to be true: when
an application has extra_src_dirs entries (or additional directories or
files) to conditionally compile under some profiles, it may start
depending on another top-level application dedicated to that profile for
include files.
However, such an app will never make it to production and neither will
the compilation artifacts that create the dependency. Under that
scenario, current rebar3 is unusable.
This patch makes it so that the compilation provider instead changes the
logic for top-level apps: rather than copying their directories one by
one and compiling them in order, it:
1. copies all top-level apps to the build directory so the files are in
their proper locations
2. adds the top-level apps to the path (after the global hooks have run,
so the existing scope and env has not changed)
3. runs the compilation as usual.
Fixes#1651
When a global plugin is used both locally and within the project, there
are cases when the rebar3 program will hard crash (killed in do_boot).
This has been traced to plugin-handling in compilation, where the same
code path may be purged twice in a row without further reloading for the
compile operation.
This of course yields the result where the code handling on the VM kills
all processes holding references to the module in memory, in this case
the rebar3 process itself.
By deduplicating the paths first, we ensure at most one purge before
reloading plugins and paths, and this prevents a hard crash.
Rather than trying one method and then the other, allow the caller to
specify the encoding of the expected file. All other schemes are risky
and won't work well.
Rollback the function's default interface to the binary format in case
any plugin used it for non-unicode content, preserving backwards compat.
The parsing functions were used inconsistently, and the returned values
were bad in some clauses; things only worked because they are never used
within rebar3. This ensures the return types are consistent on all
clauses.
The previous patch at #7c959cc fixed the usage of duplicate values
for directories through relative paths, but mistakenly went overboard
and dropped the `./` path, which is still fairly common. Similarly for
`../".
The code is modified to special-case such values and keep the code
working.
When fetching src_dir values, some relative paths can be inserted. When
deduplicating the paths on the fetch, this fact means that logically
duplicate (but literally different) directories can be returned at once.
By normalizing the names, duplication bugs can be resolved.
Specifically, this impacts profiles. It appears that relx as a whole
requires its configuration to be merged in one tuple order (New takes
precedence over Old), whereas the overlays require the opposite (Old
takes precedence over New) since the operation order on disk is
important to work well.
This patch reorders overlay values such that the overlay of a profile
takes place *after* the basic overlay, ensuring that the profile actions
take place after the basic ones; this allows profiles to properly
overwrite files as expected (see #1609)
This is done while adequately maintaining the order of operations that
were required as part of #1563
Overlay vars of profiles are also checked to be working fine, along with
a test.
This fixes#1247 and #1609
When a config file exists at the root of a project, defines a given
configuration value for a given profile, and that a sub-application
(umbrella app) also has the same profile defined with the same key (but
different values), the configuration values of the sub-application's
profile would get silently dropped.
The problem being that when the function to merge profiles is applied
recursively, it is applied to each profile (so it will merge on the keys
test, prod, etc.) rather than to each of the values of each profile.
This patch reworks the profile merging so that the current behaviour is
respected overall (a profile cannot be cancelled by a subdep's
non-existant profile since its value should have been ignored), but
ensures that sub-deps' profiles are otherwise applied recursively with
the proper rules:
- dependencies favor prior values
- plugins favor new values
- erl_first_files combine the lists
- relx uses the tuple merge algorithm
- erl_opts has its own custom merge as well
- otherwise the new value takes precedence
A test has also been added.
There is a risk of breakage in some applications that may have relied on
the buggy behaviour to work, though at this time we are aware of none of
them.
Andrey Kanyuka
Fred Hebert
Tristan Sloughter
alisdair sullivan
Sean Cribbs
Jimmy Zöger
suexcxine
getong
patrick cieplak