Fixes a known issue (cshim requires LTO) in
#1934.

…2014)

`pgrx` is switched from "C" to "C-unwind" in #1934 so it's fine to emit
an error here

Without this error, someone who updates dependencies without reading
changelog, would break extensions silently, if their only usage of
`#[pg_guard]` is `_PG_init`. It's bad.

Notes:
1. postgres must run without administrator privileges, so switch calling
`postgres` directly to starting postgres by `pg_ctl` in testing
2. Add Windows CI
3. known issue: `pgrx-cshim.c` is compiled with `-flto`, because,
without this flag, `pgrx_embed.exe` is linked to `postgres.exe`
4. known issue: `cargo-pgrx` downloads, instead of building, postgres
binaries from EDB website
5. known issue: `cargo pgrx run` and `cargo pgrx test` always print logs
for `pg_ctl start` on Windows (pipes generated by
`std::process::Command` are leaked to postgres, a command with
`Stdio::piped()` hangs, so we use `Stdio::inherit()` on Windows)
6. breaking change: this pull request fixes `PgLwLock` and `PgAtomic`
but introduces a breaking change about `PgLwLock` and `PgAtomic`: name
must be provided at `PgLwLock/PgAtomic::new`, `PgLwLock::from_named` is
removed and a parameter of `PgLwLock::attach` changes
7. breaking change: this pull request makes everything `C-unwind` and
it's necessary for downstream to switch from `C` to `C-unwind`
8. regression: `cargo-pgrx` users need to execute `sudo sysctl
fs.protected_fifos=0` before using `--runas` on Linux

…oundation#2006)

Fixes a known issue (cshim requires LTO) in
pgcentralfoundation#1934.

…gcentralfoundation#2014)

`pgrx` is switched from "C" to "C-unwind" in pgcentralfoundation#1934 so it's fine to emit
an error here

Without this error, someone who updates dependencies without reading
changelog, would break extensions silently, if their only usage of
`#[pg_guard]` is `_PG_init`. It's bad.

Notes:
1. postgres must run without administrator privileges, so switch calling
`postgres` directly to starting postgres by `pg_ctl` in testing
2. Add Windows CI
3. known issue: `pgrx-cshim.c` is compiled with `-flto`, because,
without this flag, `pgrx_embed.exe` is linked to `postgres.exe`
4. known issue: `cargo-pgrx` downloads, instead of building, postgres
binaries from EDB website
5. known issue: `cargo pgrx run` and `cargo pgrx test` always print logs
for `pg_ctl start` on Windows (pipes generated by
`std::process::Command` are leaked to postgres, a command with
`Stdio::piped()` hangs, so we use `Stdio::inherit()` on Windows)
6. breaking change: this pull request fixes `PgLwLock` and `PgAtomic`
but introduces a breaking change about `PgLwLock` and `PgAtomic`: name
must be provided at `PgLwLock/PgAtomic::new`, `PgLwLock::from_named` is
removed and a parameter of `PgLwLock::attach` changes
7. breaking change: this pull request makes everything `C-unwind` and
it's necessary for downstream to switch from `C` to `C-unwind`
8. regression: `cargo-pgrx` users need to execute `sudo sysctl
fs.protected_fifos=0` before using `--runas` on Linux

…oundation#2006)

Fixes a known issue (cshim requires LTO) in
pgcentralfoundation#1934.

…gcentralfoundation#2014)

`pgrx` is switched from "C" to "C-unwind" in pgcentralfoundation#1934 so it's fine to emit
an error here

Without this error, someone who updates dependencies without reading
changelog, would break extensions silently, if their only usage of
`#[pg_guard]` is `_PG_init`. It's bad.