either /ipfs or /ipns URIs can be used to address a
store.

To change the default values (127.0.0.1, 5001) for the
local API `host` and `port` are used in the query such
as
/ipfs/QmUNLLsPACCz1vLxQVkXqqLX5R1X345qqfHbsf67hvA3Nn?host=192.168.42.1&port=1111

If a gateway should be used instead of a local daemon
`use_gateway` is set to 1.
If the another gateway should be used, `gateway` can
be changed.

In combination:

/ipns/cache.example.com?gateway=https://gw.example.com

Build fails as:

$ make
  CXX    src/libmain/stack.o
src/libmain/stack.cc: In function 'void nix::sigsegvHandler(int, siginfo_t*, void*)':
src/libmain/stack.cc:21:21: error: 'ucontext' was not declared in this scope
     sp = (char *) ((ucontext *) ctx)->uc_mcontext.gregs[REG_RSP];
                     ^~~~~~~~
src/libmain/stack.cc:21:21: note: suggested alternative: 'ucontext_t'
     sp = (char *) ((ucontext *) ctx)->uc_mcontext.gregs[REG_RSP];
                     ^~~~~~~~
                     ucontext_t

It's caused by upstream rename:
https://sourceware.org/git/?p=glibc.git;a=commitdiff;h=251287734e89a52da3db682a8241eb6bccc050c9

which basically changes
    typedef struct ucontext {} ucontext_t;
to
    typedef struct ucontext_t {} ucontext_t;

The change uses ucontext_t.

Signed-off-by: Sergei Trofimovich <slyfox@gentoo.org>

…_binary_cache

…_binary_cache

…ache

This adds a post bool that can be used to POST content instead of GET
or PUT it.

need to use a multipart form to correctly insert the data. this
mirrors how curl -Ffile=asdf works.

This makes a number of changes:

- Use ipfs:// and ipns:// for URI, this is a proper URI and now
documented by ipfs
- Remove ipfs.hh, move everything into store
- Store daemonUri, which is the, replacing individual API settings
- Remove support for gateways
- Remove nar info cache - ipfs should do all of the caching for us
- Only retry getFile once - if it doesn’t work the ipfs daemon is
probably down
- Use /object/stat to check if file exists instead of cat

To check if IPFS daemon works, we need to do a test that should alway
work. In the future, we may want to have a min IPFS daemon that we
support, but for now any version will work.

This is okay for /ipns/ because we can always mutate it, but it’s a
hard error for ipfs.

This is really slow, but is useful to test things. It does a few API
calls to work:

- /api/v0/add : to add the new file
- /api/v0/object/patch/add-link : to insert the file into unixfs
- /api/v0/name/publish : to publish the new ipfs object to ipns

The last step is the slow part since we need to wait for it to clear.

unfortunately we don’t have a good way to handle this

for now, avoid rethrowing and just accept 500

this doesn’t wait for the final response, but updates ipns internally.

To avoid race conditions while doing /object/patch/add-link &
/name/publish, we need to lock some state. This adds a bool for
inProgressUpsert so that we can know when other threads are also
trying to publish to IPNS.

this is called at the end of a sequence of addToStores calls to tell
the store that it can now commit the data.

Hopefully we can instead pass arguments like Hydra does, soon.

(cherry picked from commit 4ced63c)

This uses the Sync primitive to store an ipfsPath that is written to
as we add stuff to IPNS. IPNS is still optional. When finished, we do
a publish on the ipnsPath.

…o ipfs-binary-cache-with-upload

… into ipfs-binary-cache-with-upload-sync

It isn't very safe, so best to do it once right after the condition. If
only we had real pattern matching...

…load-sync

Update ipfs binary cache with upload support (sync version)

…load

Update ipfs binary cache with upload support

…-develop

WIP: Get IPFS store working with StoreConfig refactor

…-develop

…-develop