Hi Punters! It’s been a while.

Following the success of the T-Shirts last year, Shuhei Takahashi has kindly designed a new ICFP Programming Contest inspired by our Lambda Map!

None of us profit off these T-Shirts; they are here purely to serve as a memento.

You can find the design on:

• Cafepress
• TMIX

You can also print your own T-Shirt if you like, using the PDF here. The image has been released under CC-BY 4.0.

Note that it contains white text, so is unsuitable for inkjet printing. The design works best on a black T-Shirt. Finally, please do make sure you check the size chart when choosing the size, as the online retailers do vary!

Thanks Shuhei – and we hope you enjoy the T-Shirts!

The source code for the lambda punter server is now available here.

The winners of the 2017 ICFP contest were announced on Tuesday evening at ICFP. Unfortunately, due to a misunderstanding, the presentation was not recorded. (A video release form was signed, but the A/V people thought it was an administrative presentation that didn’t need to be streamed or recorded!)

You can, however, watch the replays of all the games on PuntTV, and the full results are detailed below.

Ligntning contest round 3

A total of 16 teams made it through to round 3 of the lightning contest. We evaluated the punters on two large maps with the futures extension enabled.

• Edinburgh
• ICFP coauthors

The Edinburgh map is the largest one used for evaluation. It has 10000 rivers and 32 mines.

The ICFP coauthors map includes one site for each author of an ICFP paper who has coauthored a paper with another ICFP author. The rivers are the coauthor connections. The one mine in the large connected component represents Simon Peyton Jones.

Before running the evaluation we eliminated all punters who scored 0 on the Edinburgh map - due to timing out. This left us with 8 remaining teams. We played all teams together on both maps

Following a discussion on the mailing list we adopted the methodology in the task description rather than the one we had used in the previous rounds. The difference is that we compute ranking points across all games rather than computing ranking points by first aggregating the score across all games in a group. (The current approac is a bit more fine-grained, but has little effect on the final results and no effect on the winners.)

Here are the final results.

The $ound of .\ and fixstars both did very well on the Edinburgh map, scoring an order of magnitude more than other teams. However, fixstars had problems with the coauthor map and so The $ound of .\ are a long way ahead.

The winners of the prize for 1st place in the lightning contest ($500) are The $ound of ..

We officially declare that OCaml is very suitable for rapid prototyping.

Full contest round 3

A total of 32 teams made it through to round 3 of the lightning contest. We evaluated the punters on four large maps.

• ICFP coauthors (futures+splurges)
• Vancouver (futures+splurges+options)
• Oxford (futures+splurges+options)
• Edinburgh (futures+splurges+options)

The ICFP coauthor and Edinburgh maps are the same ones we used in the lightning contest. Vancouver is a map with 3601 rivers and 4 mines. Oxford is a map with 10000 rivers and 3 mines.

Before running the evaluation we again eliminated all punters who scored 0 on the Edinburgh map (this time with all extensions enabled)

• due to timing out. This left us with 17 remaining teams. We played all teams together on all maps.

Here are the final results.

Frictionless Bananas always did well, coming first on the Edinburgh and ICFP coauthor maps and second on the other two, which were more resource-constrained. GennAI won the Vancouver map, just, and Adlersprung convincingly won the Oxford map.

The prize for second place ($500) goes to DiamondPrincess.

We officially declare that C++ is a fine tool for many applications.

The prize for first place ($1000) goes to Fricitionless Bananas.

We officially declare that C++ is the programming language of choice for discriminating hackers.

(The highest placed entry written using a genuine functional programming language was from The Blind Hen, who came fourth and who used OCaml.)

Judges prize

The winners of the judges prize ($500) are powder for providing us with the junction map which we used in the second round of both contests. The powder team asked us to point out that they are the developers of “The Powder Toy” game.

We officially declare that powder is an extremely cool bunch of hackers.

The results of the second round of the full contest are now available. You can watch the replays of all the games on PuntTV.

Results

A total of 60 teams took part in the second round. The following maps were used with a selection of extension combinations always including futures.

• junction (futures+splurges, futures+options)
• gothenburg (futures+options)
• nara (futures)
• tube (futures+splurges, futures+splurges+options)

For each of the 6 map and extension combinations, we assigned groups of 16 teams (each group playing 16 times in order to allow every team to play in every position).

The median number of points is 50. Any team with fewer points is now eliminated.

The results of round 3, including the winners, will be announced at ICFP on Tuesday.

Patched punter contest (unofficial)

A number of teams have asked for more details on why their punters have been failing. In some cases this has helped them to identify small patches to make them work with our test framework. Examples of such patches include the word “Maybe” and the line “fflush(stdout);”.

Some teams have asked if we can try out their patched punters. If you can provide us with a small patch and build instructions that we can follow easily then we will be happy to run your patched punters in a special round in which we will also pit them against the eventual winners of the full contest. As we are quite busy just now, the actual evaluation may happen after the ICFP conference.

The results of the first round of the full contest are now available. Their were 146 entries in total.

You can watch the replays of all the games on PuntTV.

Bad punters

As with the lightning contest, we do not include entries that failed to score on our simplest test map. In total, 26 entries failed to score:

• ゲームセンターYAGI
• aelburicus
• ball.icfp2017
• Canivsar
• chirimenjako
• Dragons
• frosch03
• Get Counterstrike and Race for the Galaxy at http://store.steampowered.com/!
• IKS
• InforFun
• Junk Food
• KMCDragon
• lambda-llama
• Lambda Riot
• Leonardone @ NEETSDKASU
• Lonely Punter
• masakt
• MaxInt
• Milky Holmes
• OtakuPunter
• paiv
• Parse error
• sonna*baka*na
• VorpalSwords
• Xiphon and the physicists
• YukashitaOu

The most common problem seems to be that the protocol is not implemented correctly — resulting in some kind of exception being raised. Several entries fail to include the message size in their messages. There are various problems with managing the serialised game state. Some entries expect the server to send JSON state in the setup message matching a schema that is only defined by their clients. Another problem is relying on a particular order for fields of JSON objects that did not match the one of our testing framework. Some entries rely on writing state to the file system (explicitly disallowed by the specification). At least one entry always attempts to make illegal moves.

Results

A total of 120 teams took part in the first round. We used the same methodology as for the first round of the lightning contest, but with one fewer map (we only used one of the randomly generated maps compared with two in the first round of the lightning contest).

The maps are here:

• circle
• lambda
• randomMedium
• triangle

For each of the 8 possible combinations of extensions and each of the 4 maps we assigned groups of 4 teams (each group playing 4 times in order to allow every team to play in every position). This gave rise to a total of 3840 games.

The results are as follows.

The median number of points is 81.5. Any team with fewer points is now eliminated.

We did not manage to download the DrunkAlexSh full submission because it was not properly shared with icfpcontest2017@gmail.com, so their entry is the same as the one used for the lightning contest. We do not know whether this failure has anything to do with their state of inebriation at the time.

A number of other submissions had similar problems, but we think we managed to resolve them all. At least one team deliberately resubmitted their lightning entry to the full contest (which we have no problem with).

As Jeremy Sawicki helpfully pointed out on the mailing list, there were some glitches in the move order for the first round of lightning results.

Lightning contest round 1 take two

We have now re-run the lightning round 1 tests and fortunately doing so did not have a huge impact on the result (most teams that did well the first time also did well the second time round).

The alternative results, which we believe fairly assign move order, are as follows.

The median number of points is 26. The important differences are that 2 teams no longer make the cut

• DiamondPrincess
• master_thesis

and 4 teams who did not do so previously now do:

• Enoch Root
• Sampou
• Sirius Cybernetics Corporation
• Skobochka

Given that we had already announced that DiamondPrincess and master_thesis had made it to round 2, we decided it not to eliminate them at this stage.

Repechage

With a bit of effort we managed to free some of the punters that had previously had their poles stuck in the mud. The 9 punters that we managed to free are the following:

• Adlersprung
• A Storm Of Minds
• chirimenjako
• code-o-matic
• LILiK
• powder
• ?!?
• Sideways Spider
• Turing Machinists

Some of these were failing to find dynamic libraries. A few more required a longer setup time than we were previously allowing. A few more do not deal correctly with redundant meta data that we were originally passing. Finally, some of them have not quite implemented the protocol correctly in that they depend on the input stream being closed after the server has sent its messages. (Part of the reason we found these fixes is because we tried a lot harder with the full contest entries and some of those attempts revealed corresponding fixes to the lightning round entries.)

We are sorry that a few punters still have their poles stuck in the mud or have sunk completely. There is only so much we can do.

One entry, chirimenjako, cheats in that it relies on writing to the file system. We decided to let this go for the lightning round as we are being a bit lax, the team were open about not managing to get serialisation of state working with lamduct, and the files they generate are small. But they will be disqualified from the full contest. (We will accommodate first-run caching and be more lenient towards punters who simply write a log file - in the latter case we will replace the log file with a soft link to /dev/null.)

With all of these newly free punters, and those who wrongly missed out in our first attempt at round 1, we decided to hold a repechage. The rules and maps are the same as for round 1 proper, but only the 4 punters who lost out to the bad move order and the 9 who got their poles stuck in the mud contested it.

The results are here.

The median number of points is 25. The teams with fewer points are eliminated.

Lightning round 2

Having run the repechage we were then in a position to run round 2 with all of those punters still in the competition.

For round 2, we used the same scheme as for round one, except the maps were a little larger (including one submitted by a participant), and we had 8 players per map. We used the following four maps.

• boston
• junction
• star
• tube

As each punter contested each map with and without futures, the minimum number of points is 8 and the maximum number of points is 64.

The results are as follows.

The median number of points this time is 37. All punters with fewer points are eliminated for the third round. The winner of the third (and final) round will be the winner of the lightning contest and will be announced at ICFP.

This time, a number of punters struggled with the larger maps and timed out. The lowest scoring punters had particular problems here. However, even some of the higher scoring punters timed out occasionally on the larger maps.

PuntTV

As usual you can watch replays of all the games on PuntTV.

We have now run the first elimination round for the lightning contest. There were 80 entries in total.

PuntTV

You can now watch replays of all the test games on PuntTV.

Bad punters

As a preprocessing step we removed all entries that were incapable of scoring anything at all on our simplest test map: lambda.json, even if they moved first.

Some teams did not implement the handshake. Given that it was not present in the original (slightly buggy) version of the spec, we wrote a small wrapper that still enabled those teams to compete.

The following 28 teams were unable to score anything on the lambda map.

• Adlersprung
• A Storm Of Minds
• chirimenjako
• code-o-matic
• d4o
• DNIWE :: a
• Get Counterstrike and Race for the Galaxy at http://store.steampowered.com/!
• Goto11
• JODY HiGHROLLERS
• Junk Food
• kstm.org
• lambda-llama
• Leonardone @ NEETSDKASU
• LILiK
• masakt
• MIPT Lambda
• ome
• perpertuum mobile
• powder
• ?!?
• Sideways Spider
• sonna*baka*na
• Sunspear
• TBD
• The Cat is #1!!
• The Flux Ambassadors
• The Pragmatic Russians
• Turing Machinists

They are now eliminated from the lightning round.

One team (cashto) was able to score on the lambda map when futures are disabled but not when enabled. Another team (negainoido) was able to score on the lambda map when futures were enabled but not when disabled.

Calculating the scores

For the first round we used five small maps (with between 60 and 187 rivers) both with and without futures enabled. Every game was played with 4 teams.

For each map a random assignment was performed and for each group of four players, four games were played on that map (in order to allow each player to start first, second, third, and fourth). Thus each team (with the exception of cashto and negainoido) played a total of 40 games.

For each group, the sum of the scores for each player in that group over the four games was calculated in order to obtain a ranking. Points were then awarded based on the ranking (1st: 4 points, 2nd: 3 points, 3rd: 2 points, 4th: 1 point).

The total number of points for each team was calculated by taking the sum of the points for each group they played in. So the maximum possible number of points was 40 and the minimum number was 10 (with the exception of cashto and negainoido, who only played in five groups each).

The results

The median number of points is 25.5. Any team with a points total below 25.5 is now eliminated from the lightning contest.

Interestingly, there is not a direct correlation between aggregate score and aggregate points. For instance, the team with the highest aggregate score (fixstars) is the team with the 8th highest aggregate points. Team shinh11 deserves special mention for being the biggest (and worst) gamblers.

(photograph from fisuk.org)

It is time to hop off the punt and go ashore. The contest ended a little over an hour ago. We would like to thank all contestants for punting with us. We hope you enjoyed punting on the ICFP contest rivers.

Submission

Please make sure your latest submission is accessible by us, e.g. visit the Google Drive shareable link in incognito mode.

Quite a few contestants have told us that they forgot to type their shareable link along with the MD5 checksum into the submission system. In case you find yourself in this situation then you can still register your submission via the submission system. Note: We only accept this submission if the timestamp on your Google Drive submission is prior to the official submission deadline.

Evaluation

We will post further information regarding the evaluation of submissions in the coming weeks. Stay tuned!

Languages powering the contest

For this contest we have used a variety of programming languages. The servers and tools were written in the following languages

• Bash
• Haskell
• Java
• Javascript
• Latex
• Links
• jq
• OCaml
• Python
• Ruby
• Rust

Yet again, thank you punting with us!

Hello again punters! We hope you are enjoying the last part of the contest.

The contest will be ending in three hours time. Please remember to share your entries with icfpcontest2017@gmail.com on Google Drive and to register here.

Good luck!

Hello punters!

Tony Hoare refers to null pointers as his billion-dollar mistake. Option types, popularised by functional programming languages, provide a much cleaner alternative.

This extension has nothing whatsoever to do with null pointers or option types!

New extension: Options

This extension enables options, which allow a punter to use a river claimed by another punter. Every punter is allowed to buy n options during the game where n is the number of mines. Each river claimed by a punter has exactly one option which might be bought by a second punter.

The rules for scoring are unaffected – but sharing a route with another punter might be a good option from time to time.

You can find the specification here.

Happy punting, and good luck!

Hello again punters! We hope you are enjoying the contest.

New extension: Splurges

In 1984, Philip Wadler posited that listlessness is better than laziness. We would like you to help us investigate this claim!

This extension enables splurges, which allow a punter to be lazy (pass!) for n turns, and then claim a list of n + 1 contiguous rivers at once.

The rules for scoring are unaffected — but being lazy may improve your chances of building a good route!

You can find the specification here.

Lambda Duct v0.3

Earlier, we also updated Lambda Duct to version 0.3. Lambda Duct allows you to test your offline mode by relaying your moves to the game servers.

The major change from v0.2 is that v0.3 uses blocking I/O instead of non-blocking I/O.

Good luck, and happy punting!

Hello punters!

We hope you are getting on well.

Due to the webserver glitches at the start of the contest, we are extending the lightning round deadline by 1 hour. Therefore, the new deadline will be: 13:00 UTC

To submit for the lightning division register here, package your submission into a single .tar.gz file with the registration code as file name, upload it to GoogleDrive, and share the entry with icfpcontest2017lightning@gmail.com. More details in section 7 of the problem description.

Good luck!

We have released the initial version of Lambda Duct – an offline mode simulator for Lambda Punter. Lambda Duct acts as a bridge between a client in offline mode and an online game server.

Please refer to the Lambda Duct repository on GitHub for instructions on how to install and use it.

TL;DR

Install Lambda Duct via OPAM

$ opam pin add lambda-duct https://github.com/icfpcontest2017/lambda-duct.git

Then supply the name of your client program (configured to run in offline mode) and the port of the game you want to connect to

$ lamduct --game-port <port> ./<name of client program>

Note the ./ in front of the client program name.

Happy punting!

The Futures are here!

We are releasing the first extension of the contest: Futures. Futures are bets on reaching a particular site from a mine and are well rewarded if fulfilled. But beware because failure can be costly …

The Futures specification is here.

We also made modifications to the problem description bumping it to version 1.3.

Changes

• Fixed typo in the type of the contents of the “timeout” message: Float rather than GameState.

• Clarified which moves are sent along with a stop message.

• Clarified that the size header in a message is a string.

• Clarified that the VM will be used for the evaluation.

We updated the task description to version 1.2.

Biggest change is an update of the protocol to clarify how timeouts are measured.

Changes

• Updated protocol to include handshakes in offline mode and to specify more precisely where timeouts occur.

• Clarification on the game outcome when multiple punters score equally.

• Clarification over the treatment of illegal moves by the server.

• Clarification over the maximum length of messages.

• Clarified meaning of disjoint union operator.

Hello punters!

We hope you’re getting on well with punting so far.

We apologise for the initial issues with the web server being overloaded when the task was announced – we’d thought that the university servers would be able to take more punishment! We have now moved the website over to GitHub so this shouldn’t happen again.

There have also been some teething problems with PuntFX, but we are working these out. These will not affect punters connecting directly to the test servers.

We will take the servers offline in 15 minutes for maintenance; we expect downtime to be around 10 minutes.

The task description is now available here. Good luck!

VM update

We have uploaded an updated version of the VM. A number of other requests have come in and we will upload a further update early next week.

We recognise that some participants may only realise that they want to use a particular tool or library once they know what the task is. That should be fine providing the requested package is reasonably easy to install on the VM and you include installation instructions along with your submission.

We will give full details of the submission process along with the task description next Friday.

The contest begins in two weeks time. Here is some preliminary information about the rules.

Teams

Each participant must be a member of exactly one team. Teams may not divide or collaborate with one another once the contest has begun. There is no limit to the number of members of a single team.

Intellectual property

Contest participants retain ownership of all intellectual property rights in and to any submitted source codes, custom tools, and related materials (“Submissions”) that participants had before submission. As a condition of entry, participants grant the judges a non-exclusive, perpetual, irrevocable, worldwide, royalty-free license to use, reproduce, publish, distribute, publicly perform, and publicly display the Submissions for the purposes of allowing the judges to test and evaluate the Submissions for purposes of the contest.

Welcome to the official website for the ICFP Programming Contest 2017, the 20th instance of the annual programming contest series sponsored by The ACM SIGPLAN International Conference on Functional Programming.

This year’s contest is brought to you by LFCS at the University of Edinburgh. It will be held for 72 hours, starting from 4th August 2017 12:00 UTC and 7th August 2017 12:00 UTC.

Task

The task will be published on this website when the contest starts. Solutions must be submitted online before the contest ends. Details of the submission procedure will be announced along with the contest task.

The contest is open to everyone except for contest organisers, their families, and members of their institutions. Neither advance registration nor entry fee is required.

Programming languages

Any programming languages may be used as long as the submitted program can be run by the judges on a standard Linux environment with external network access disabled.

Environment

The judging environment is a virtual machine running a recent version of Debian with a selection of programming languages installed. It is available as a Virtual Box image here.

For further details take a look at the readme, which contains instructions and details the default languages and environments that are installed.

Please let us know before the contest if you would like us to install anything else.

Prizes

There will be prizes for the first and second place teams, the team winning the (24 hour) lightning division, and a discretionary judges’ prize. The winners will be announced at ICFP 2017 in Oxford, UK, on Tuesday 5th September 2017.

In addition, the organisers will declare during the conference that:

• the first place team’s language is “the programming language of choice for discriminating hackers”,
• the second place team’s language is “a fine tool for many applications”,
• the winning lightning division team’s language is “very suitable for rapid prototyping”, and
• the team winning the judges’ prize is “an extremely cool bunch of hackers”.

Updates

We will post updates to this website. You can also:

• follow us on Twitter (@ICFPContest2017),
• subscribe to the mailing list (the list address is icfp-contest2017 [at] inf.ed.ac.uk), and
• join the IRC channel, #icfp-contest, on irc.freenode.net.