I don't see how that issue would stop you from using Anarki. And it appears news.arc was modified to accommodate the change.

I have to say, I don't like that change in making templates into first class objects. It would have been better to add something like 'inst-with' to extend inst.

If that change is a show stopper for you, then I would just fork anarki and eliminate the template changes, or fork Arc 3.1 and add the file upload change.

Forking Anarki also means maintaining the fork, which means one more thing I have to worry about and probably break. I'd rather only add what I need: the ability to have file uploads to Arc 3.1. Uploading files through HTTP is needed regardless of the storage behind the application (filesystem vs database).

MongoDB: http://docs.mongodb.org/ecosystem/tools/http-interfaces/

Datomic: http://blog.datomic.com/2012/09/rest-api.html

OrientDB: https://github.com/orientechnologies/orientdb/wiki/OrientDB-...

MySQL http://code.nytimes.com/projects/dbslayer

Riak: http://docs.basho.com/riak/latest/dev/references/http/

And so on.

Of course Anarki also has HTTP Get/Post functionality where Arc 3.1 does not:

https://github.com/arclanguage/anarki/blob/master/lib/http.a...

You would also need to parse the HTTP results.

If you're dead-set on Postgres you could do something like this: http://rny.io/nginx/postgresql/2013/07/26/simple-api-with-ng... Though this seems a little a hackish to me.

Could I ask why you wouldn't go back to saving data in files?

It's a clean, fast solution if one doesn't need joins. Using SQL requires building up strings, or putting together an ORM. Better not go down that path if you don't need them.

It's not the file structure itself that's the issue, it's mechanisms built around accessing the file(s). For a SMALL HN clone it's probably OK, but if the requirements change even a little you're in for some trouble.

Food for thought: Once your app data can't fit into memory you're going to need to move to a real database or spend big bucks on more hardware. You could do what HN does and purge older data (reloading only when required), but still if your app design requires regular access/writes to older data and it doesn't all fit into memory then your hardware is going to start to thrash (HN had these type of problems from what I read). And with HN, since each record is stored in it's own file, it's going to be exponentially bad.

Also, it's pretty useful to have your dbs and applications running on different servers. Doing this provides other applications running elsewhere to access the data too. So once again it all boils down to your design and expectations, which is why I originally asked if you are just running a HN clone.

FYI I just posted a FleetDB v2 link on the Arc forum if you want to check that option out.

Access to older data is equally a problem for the filesystem and a database. Databases don't get away with it. Once the app data can't fit into memory databases thrash on disk too.

The only argument I see is that save-table saves an entire record in a single file. Wanting access to a specific field of that field means loading the entire record. A lot of databases function this way too, but in general they don't have to. They could bring in memory individual fields. You could claim that as the reason for filesystem access to be slow.

But even then, save-table could be modified to save each field in a separate file.

No problem. Note though that FleetDB is all in memory too. The advantage is that you can:

  1. Put the database on its' own server with dedicated RAM. 
  2. Compose and re-use queries.
  3. Have index support, built in, without code writing.
  4. Have robust/high performance concurrency semantics, not just atomic file writes.

You're correct in that they too will do seeks to disk, but nowhere close to the same amount when compared to handling each record in it's own file. Just try loading a million files vs 1 file with a million records.

> But even then, save-table could be modified to save each field in a separate file.

Which will exponentially increase the number of disk seeks.

As I stated before it really depends upon your design. Ask yourself: Are you going to have say a million records where you have to run a query againts all records? Would indexes help with that query on that data? Do you really want to code each index into a table then change all your application code to use them? Or would you rather instruct your database to add an index and discover the existing queries already take advantage of them? Note that the last time I checked, HN off loaded that kind of heavy lifting to a third party service[1] that uses[2], you guessed it, a database!

I am not a database expert and I don't want to convince you to use tool that you don't need to use. I'm just saying I see enough advantages in using a database that I don't plan on using file writes.

It seems to me, the original problem was this: One defines a template with a non-nil value as the default for a fields argument, then upon instantiating the template one wants to pass in a nil value to eliminate that fields default non-nil value from being part of the resulting table.

By making templates first class objects you have permitted these over-rides to occur, but at the same time you have introduced an additional layer of complexity for every user of templates (and in some cases even tables), when they don't care about that specific circumstance.

So at first glance I can't help wonder why not just re-write templates in a manner that eliminates poor handling non-nil values. Then we could have maintained the inst function while also allowing for other methods to be created, such as an 'inst-with' that allows nil values to over-ride defaults (or just have inst allow for over-rides).

I'm not sure how templates are written, but I imagine they make heavy use of tables, which is where the nil pruning is probably occurring. So templates would instead need to use association lists up until the output stage where data then gets pushed into a table.

Or, if instead, you want tables to support nil values, then that's also a change worth considering, but I see the introduction of a whole new type as a means to solve a certain circumstance as overkill with downside effects.

The problem with not supporting nil values in templates -- and therefore in the data models of any webapp -- is that you can't save booleans as booleans. That seems ugly.

Adding new names to paper over problems with existing names seems ugly as well.

Your idea of questioning the design of tables themselves is interesting. I hadn't considered that. A persistent data structure seemed like a separate beast, distinct from a table. But perhaps all tables should be easy to persist. Hmm.

---

My attitude towards anarki has been that it's trying to be the core for a hundred-year language, so it should get the core right without regard to how things were before. So we should avoid unseemly hacks for things that seem simple -- like storing a user pref that's a boolean -- without regard to how commonly they happen, or how easy workarounds are to find.

Hmm. Well I see that as a separate issue. I think nil and boolean false are very different things. nil means no value or false while false means a false value. In fact I was recently going to suggest that arc should support proper booleans and that arc tables should also store both boolean values, while maintaining it's nil pruning feature. Which really stems from wanting to easily transform to/from json or edn like formats. Currently one has to fudge arc's write-json by passing around symbols for 'true / 'false.

> how would you design templates if you were doing so from scratch?

Well that's interesting that you ask, because I did just that when I implemented templates in Clojure. Of course Clojure supports true boolean values and maps can hold nil values, so it was trivial and very useful.

My Clojure templates are on steroids though. Not only do they match most of the features in arc, but they are cumulative, and accept anonymous functions as field values. The values can also refer to other fields inputs or results.

So for example I just did this one:

  => (deftem :article
      :id        #(UUID) 
      :published true       
      :text      nil 
      :msec      #(msec)
      :seconds   #(seconds (% :msec))
      :date      #(format-date (date (% :msec)) "yyyy/MM/dd HH:mm:ss"))
  
  #result 
  
  => (invoke :article :text "cool" :published nil)
  
    {:id #uuid "08692e96-de87-49aa-9ad1-e33bcd61e712", 
     :text "cool", 
     :msec 1382644173685, 
     :seconds 1382644173, 
     :date "2013/10/24 15:49:33"}

  1. In bad shape (wrote it early on) 
  2. Includes partial features not fully implemented.
  3. Has references to functions I can not split out without  
     creating a bunch of work.
  4. Includes features you would not care for (datomicish).
  5. Has oddities that make you wonder "why like that",
     until you realize you can pass in say a map of args   
     instead. 

  (def index* (ref (hash-map)))
  (def templates* (ref (hash-map)))
  (def mutes* (ref (hash-map)))
  (def selfs* (ref (hash-map)))

  (defmacro deftem [name & fields]
    `(let [tem#   (quote ~name)
	   order# (evens (list ~@fields))
	   fmaps# (apply hash-map (list ~@fields))]
	(dosync (alter templates* assoc tem# fmaps#)
	        (alter index* assoc tem# order#)
	    fmaps#)))

  (defmacro defmute [name & fields]
    `(let [tem#   (quote ~name)
	  items# (list ~@fields)]
      (dosync (alter mutes* assoc tem# items#)
        items#)))

  (defmacro defself [name & fields]
   `(let [tem#   (quote ~name)
	  items# (list ~@fields)]
      (dosync (alter selfs* assoc tem# items#)
        items#)))

  (defn invoke-fields
   ([tem base fields allowables]
     (invoke-fields tem base fields allowables nil))
   ([tem base fields allowables testfn]
    (let [fks   (keys fields)
          selfs (@selfs* tem)]
      (reduce
       (fn [m k]
          (assoc m k
	        (if (detect? k fks); must use 'detect' opposed to 'find' for nil vals must be inserted.
		          (aifn (fields k)
	                  (try (it m)
	                    (catch Exception e (it)))
	                  (let [bfn (base k)]
                      (if (and (detect? k selfs)(fn? bfn))
                          (try (bfn (merge m {k it}))
	                          (catch Exception e (bfn)))
                           it)))
              (aifn (base k)
                    (try (it m)
                         (catch Exception e (it)))
                     it))))
		 (hash-map) allowables))))


  (defn invoke [tem & fields]
    (let [temx  (split-name tem)
          tem1  (first-identity temx)
	  atem? (is (last temx) "+")
	  xfn   (type-fn tem)
          temk  (xfn tem1)
	  base  (@templates* temk)
	  prox  (@mutes* temk)
	  fval  (first fields)
	  fmap  (cond (map? fval) fval ; for file loading map of saved records
	              (coll? fval) (apply hash-map fval)
	           :else (apply hash-map fields))
	  imap  (invoke-fields temk base fmap (@index* temk))]
       (reduce
	   (fn [m [k v]]
		   (if (or (missing? k base)(nil? v)(detect? k prox))
	         (dissoc m k)
		       (assoc m (if atem? (nsify temk k) k) v)))
	   	(hash-map) imap)))

  arc> (save-table (obj a 10 b 20) "tfile")
  nil

  arc> (load-tables "tfile")
  (#hash((b . 20) (a . 10)))

If you haven't made any changes to templates (created your own, modified the ones in news.arc) you should be safe.

For instance, suppose Anarki defines a new utility and uses it to simplify the implementation of 10 other utilities. (It does this in a few places.) Now suppose my Arc 3.1 code has defined a utility with exactly the same name, and running this code on Anarki causes those other 10 utilities to misbehave, thus wrecking my program. This is a case where Anarki isn't compatible with Arc 3.1, but since it's so easy for me to choose a different name for my utility, it's hardly even worth mentioning. Pretty much any substantial update to Arc would break it in exactly the same way.

There's only one difference between Arc 3.1 and Anarki that's ever gotten in my way, and that's the way Anarki has revamped the [...] syntax to support multi-argument functions. When I say [do t] or [do `(eval ',_)], Anarki treats these as 0-arity functions, and when I say [let (a . b) _ ...], Anarki chokes when trying to search the dotted list for any underscored variables. Once again, this is the kind of change that's pretty easy to work around, and I can't really say Anarki is worse for having this extra functionality.