sneeuwballen / benchpress Goto Github PK
View Code? Open in Web Editor NEWTesting and benchmarking tool for logic-related programs.
License: BSD 2-Clause "Simplified" License
Testing and benchmarking tool for logic-related programs.
License: BSD 2-Clause "Simplified" License
It is very convenient that currently benchpress records the time for each prover run, but it would be even more useful to also have an estimation of the memory used by each process. Fortunately, the GNU time utility provide such a fonctionality.
Something to that effect could be obtained by wrapping the prover calls like this:
/usr/bin/time -f "%e\n%M" -o some_tmp_file prover_cmd_to_run....
And then reading the temporary file for the output of time (%e
is for the elapsed real time in seconds, and %M
the maximum resident set size in Kbytes).
logitest-starexec
?)instead of expect = unknown
(as a fallback for problems where an expect string isn't found), a useful feature for benchmarks is to force the expect field to be unknown (or even just empty). In this case we could display results instead of ok/bad…
In the summary of a prover's results for a task, there are lines for each tag that provides the number of runs that resulted in this tag; it would be nice if that number also was a link to the list of the runs with that result (which is possible thanks to the new filtering capability of the list of results).
Currently it seems that provers run by benchpress are wrapped twice in a ulimit wrapper, which is at the very least superfluous, and I suspect might be one reason why I continue to have zombie processes when running benchmarks.
Additionally, the two wrappers do not enforce the same limits, which may or may not make things even more confusing.
For the record here is the command run for a timeout of 600s and 2G memory limit, as per benchpress logs:
cmd "ulimit -t 600 -m 2000000 -Sv 16000000; ulimit -t $(( 1 + 600)) -v $(( 1200 * 2000 )); <prover_cmd>
Benchpress sometimes crashes while running on the following configuration:
(import-prelude false)
(prover
(name ae-read-status)
(cmd "zgrep :status $file")
(unknown ":status unknown")
(unknown "")
(sat ":status sat")
(unsat ":status unsat|:status valid"))
(dir
(path "tests")
(pattern ".*.ae|.*.smt2|.*.zip")
(expect (run ae-read-status)))
(prover
(name random)
(cmd "./random_solver --timelimit=$timeout $file")
(sat "^sat")
(unsat "Valid|(^unsat)")
(unknown "(I Don't Know)|(^unsat)")
(timeout "^timeout"))
where the solver random_solver
is
#!/bin/bash
choice=$((RANDOM % 5))
case $choice in
'0')
echo "unsat";;
'1')
echo "sat";;
'2')
echo "unknown";;
'3')
echo "error";;
'4')
echo "timeout";;
esac
The trace of the exception:
Sys_error("Mutex.lock: Resource deadlock avoided")
Raised by primitive operation at CCLock.with_lock in file "src/threads/CCLock.pp.ml" (inlined), line 11, characters 2-20
Called from Benchpress__Misc.synchronized in file "src/core/Misc.ml", line 15, characters 21-45
Called from Benchpress__Misc.Log_report.reporter.report.k.write_str in file "src/core/Misc.ml", line 67, characters 10-75
Called from Benchpress__Run_prover_problem.pp_result in file "src/core/Run_prover_problem.ml", line 64, characters 2-208
Called from Stdlib__Format.kfprintf.(fun) in file "format.ml", line 1356, characters 16-34
Called from CCLock.with_lock in file "src/threads/CCLock.pp.ml" (inlined), line 13, characters 12-23
Called from Benchpress__Misc.synchronized in file "src/core/Misc.ml", line 15, characters 21-45
Re-raised at CCLock.with_lock in file "src/threads/CCLock.pp.ml" (inlined), line 18, characters 4-11
Called from Benchpress__Misc.synchronized in file "src/core/Misc.ml", line 15, characters 21-45
Called from Benchpress__Run_prover_problem.pp_result_progress in file "src/core/Run_prover_problem.ml", line 75, characters 2-109
Called from Benchpress__Exec_action.Progress_run_provers.progress.object#on_res in file "src/core/Exec_action.ml", line 410, characters 27-84
Called from Benchpress__Exec_action.Exec_run_provers.run.run_prover_pb.(fun) in file "src/core/Exec_action.ml", line 279, characters 6-21
Called from Benchpress__Error.guard in file "src/core/Error.ml", line 40, characters 6-9
The output produced by the command : dune exec -- benchpress run -c config.sexp -t 5 -p random ./tests -vv
[2023-03-20T23:22:13|t0] [debug:] logs are setup [2023-03-20T23:22:13|t0] [info:] parse config files ["config.sexp"] [2023-03-20T23:22:13|t0] [info:] add-stanza (enter-file /home/tiky/programming/test3/config.sexp) [2023-03-20T23:22:13|t0] [info:] add-stanza (prover (name ae-read-status) (cmd "zgrep :status $file") (sat ":status sat") (unsat ":status unsat|:status valid") (unknown "")) [2023-03-20T23:22:13|t0] [info:] add-stanza (dir (path tests) (expect (run ae-read-status)) (pattern ".*.ae|.*.smt2|.*.zip")) [2023-03-20T23:22:13|t0] [debug:benchpress.definition] cur dir: '/home/tiky/programming/test3' [2023-03-20T23:22:13|t0] [info:] add-stanza (prover (name random) (cmd "./random_solver --timelimit=$timeout $file") (sat "^sat") (unsat "Valid|(^unsat)") (unknown "(I Don't Know)|(^unsat)") (timeout "^timeout")) [2023-03-20T23:22:13|t0] [info:benchpress.run-main] run-main.main for paths ./tests [2023-03-20T23:22:13|t0] [debug:] check prefix dir="/home/tiky/programming/test3/tests" for "/home/tiky/programming/test3/./tests" [2023-03-20T23:22:13|t0] [debug:] get-cmd-out "grep -c processor /proc/cpuinfo" [2023-03-20T23:22:13|t0] [debug:] par-map: create pool j=3 [2023-03-20T23:22:13|t3] [debug:problem] (find_expect `/home/tiky/programming/test3/./tests/AUFLIA/misc/arr2.smt2` using (run ae-read-status))… [2023-03-20T23:22:13|t3] [debug:prover] (Prover.run ae-read-status (limits (timeout 1s) (memory 200M))) [2023-03-20T23:22:13|t1] [debug:problem] (find_expect `/home/tiky/programming/test3/./tests/AUFLIA/misc/set3.smt2` using (run ae-read-status))… [2023-03-20T23:22:13|t1] [debug:prover] (Prover.run ae-read-status (limits (timeout 1s) (memory 200M))) [2023-03-20T23:22:13|t2] [debug:problem] (find_expect `/home/tiky/programming/test3/./tests/AUFLIA/misc/set1.smt2` using (run ae-read-status))… [2023-03-20T23:22:13|t2] [debug:prover] (Prover.run ae-read-status (limits (timeout 1s) (memory 200M))) [2023-03-20T23:22:13|t3] [debug:run-proc] (run.done :errcode 0 :cmd "ulimit -t 2 -Sm 200 ; zgrep :status /home/tiky/programming/test3/./tests/AUFLIA/misc/arr2.smt2" [2023-03-20T23:22:13|t3] [debug:run-proc] stdout: (set-info :status unsat)stderr:
[2023-03-20T23:22:13|t3] [debug:problem]
(find_expect/home/tiky/programming/test3/./tests/AUFLIA/misc/set2.smt2
using (run ae-read-status))…
[2023-03-20T23:22:13|t2] [debug:run-proc]
(run.done :errcode 0
:cmd "ulimit -t 2 -Sm 200 ; zgrep :status /home/tiky/programming/test3/./tests/AUFLIA/misc/set1.smt2"
[2023-03-20T23:22:13|t3] [debug:prover]
(Prover.run ae-read-status (limits
(timeout 1s)
(memory 200M)))
[2023-03-20T23:22:13|t2] [debug:run-proc]
stdout:
(set-info :status unsat)stderr:
[2023-03-20T23:22:13|t1] [debug:run-proc]
(run.done :errcode 0
:cmd "ulimit -t 2 -Sm 200 ; zgrep :status /home/tiky/programming/test3/./tests/AUFLIA/misc/set3.smt2"
[2023-03-20T23:22:13|t1] [debug:run-proc]
stdout:
(set-info :status unsat)stderr:
[2023-03-20T23:22:13|t3] [debug:run-proc]
(run.done :errcode 0
:cmd "ulimit -t 2 -Sm 200 ; zgrep :status /home/tiky/programming/test3/./tests/AUFLIA/misc/set2.smt2"
[2023-03-20T23:22:13|t3] [debug:run-proc]
stdout:
(set-info :status sat)stderr:
[2023-03-20T23:22:13|t0] [debug:] par-map: stop pool
testing with 1 provers, 4 problems…
[2023-03-20T23:22:13|t0] [debug:benchpress.runexec-action]
output database file /home/tiky/.local/share/benchpress/res-20230320T232213-9118d387-bca8-494c-8e72-3ecd4c7f25ed.sqlite
[2023-03-20T23:22:13|t0] [debug:] par-map: create pool j=1
[2023-03-20T23:22:13|t8] [info:benchpress.run-prover-pb]
running random //home/tiky/programming/test3/./tests/AUFLIA/misc/set3.smt2 (timeout 5s)...
[2023-03-20T23:22:13|t8] [debug:benchpress.run-prover-pb] proof file: None
[2023-03-20T23:22:13|t8] [debug:prover]
(Prover.run random (limits
(timeout 5s)))
[2023-03-20T23:22:13|t8] [debug:run-proc]
(run.done :errcode 0
:cmd "ulimit -t 6 ; ./random_solver --timelimit=5 /home/tiky/programming/test3/./tests/AUFLIA/misc/set3.smt2"
[2023-03-20T23:22:13|t8] [debug:run-proc] stdout:
satstderr:
[2023-03-20T23:22:13|t8] [debug:]
output for ./random_solver//home/tiky/programming/test3/./tests/AUFLIA/misc/set3.smt2:sat
, ``, errcode 0
[2023-03-20T23:22:13|t8] [info:benchpress.run-prover-pb]
running random //home/tiky/programming/test3/./tests/AUFLIA/misc/set1.smt2 (timeout 5s)...
[2023-03-20T23:22:13|t8] [debug:benchpress.run-prover-pb] proof file: None
[2023-03-20T23:22:13|t8] [debug:prover]
(Prover.run random (limits
(timeout 5s)))
Error:
Sys_error("Mutex.lock: Resource deadlock avoided")
Raised by primitive operation at CCLock.with_lock in file "src/threads/CCLock.pp.ml" (inlined), line 11, characters 2-20
Called from Benchpress__Misc.synchronized in file "src/core/Misc.ml", line 15, characters 21-45
Called from Benchpress__Misc.Log_report.reporter.report.k.write_str in file "src/core/Misc.ml", line 67, characters 10-75
Called from Benchpress__Run_prover_problem.pp_result in file "src/core/Run_prover_problem.ml", line 64, characters 2-208
Called from Stdlib__Format.kfprintf.(fun) in file "format.ml", line 1356, characters 16-34
Called from CCLock.with_lock in file "src/threads/CCLock.pp.ml" (inlined), line 13, characters 12-23
Called from Benchpress__Misc.synchronized in file "src/core/Misc.ml", line 15, characters 21-45
Re-raised at CCLock.with_lock in file "src/threads/CCLock.pp.ml" (inlined), line 18, characters 4-11
Called from Benchpress__Misc.synchronized in file "src/core/Misc.ml", line 15, characters 21-45
Called from Benchpress__Run_prover_problem.pp_result_progress in file "src/core/Run_prover_problem.ml", line 75, characters 2-109
Called from Benchpress__Exec_action.Progress_run_provers.progress.object#on_res in file "src/core/Exec_action.ml", line 410, characters 27-84
Called from Benchpress__Exec_action.Exec_run_provers.run.run_prover_pb.(fun) in file "src/core/Exec_action.ml", line 279, characters 6-21
Called from Benchpress__Error.guard in file "src/core/Error.ml", line 40, characters 6-9Context:
(running :prover random :on /home/tiky/programming/test3/./tests/AUFLIA/misc/set3.smt2 (expect: unsat))Context:
running 4 testsContext:
run prover action
(run_provers (dirs (/home/tiky/programming/test3/./tests :in (dir (path /home/tiky/programming/test3/tests) (expect (run ae-read-status)) (pattern ".*.ae|.*.smt2|.*.zip")))) (provers random) (timeout 5s) (j 1))
The state of the local opam switch I used for this issue:
# Packages matching: installed # Name # Installed # Synopsis base-bigarray base base-bytes base Bytes library distributed with the OCaml compiler base-threads base base-unix base base64 3.5.1 Base64 encoding for OCaml benchpress 0.1 pinned to version 0.1 at git+https://github.com/sneeuwballen/benchpress.git benchpress-lsp 0.1 pinned to version 0.1 at git+https://github.com/sneeuwballen/benchpress.git benchpress-server 0.1 pinned to version 0.1 at git+https://github.com/sneeuwballen/benchpress.git cmdliner 1.1.1 Declarative definition of command line interfaces for OCaml conf-gnuplot 0.1 Virtual package relying on gnuplot installation conf-pkg-config 2 Check if pkg-config is installed and create an opam switch local pkgconfig folder conf-sqlite3 1 Virtual package relying on an SQLite3 system installation containers 3.11 A modular, clean and powerful extension of the OCaml standard library containers-thread 3.11 An extension of containers for threading cppo 1.6.9 Code preprocessor like cpp for OCaml csexp 1.5.1 Parsing and printing of S-expressions in Canonical form csv 2.4 A pure OCaml library to read and write CSV files dune 3.7.0 Fast, portable, and opinionated build system dune-configurator 3.7.0 Helper library for gathering system configuration either 1.0.0 Compatibility Either module gnuplot 0.7 Simple interface to Gnuplot Gnuplot-OCaml provides a simple interface to Gnuplot from OCaml. The API supports only 2D graphs and was inspir ISO8601 0.2.6 ISO 8601 and RFC 3999 date parsing for OCaml iter 1.6 Simple abstraction over `iter` functions, intended to iterate efficiently on collections while performing some transformations jemalloc 0.2 Bindings to jemalloc mallctl api jsonrpc 1.5.0 Jsonrpc protocol implemenation linol 0.4 LSP server library logs 0.7.0 Logging infrastructure for OCaml lsp 1.5.0 LSP protocol implementation in OCaml ocaml 4.14.1 The OCaml compiler (virtual package) ocaml-config 2 OCaml Switch Configuration ocaml-syntax-shims 1.0.0 Backport new syntax to older OCaml versions ocaml-system 4.14.1 The OCaml compiler (system version, from outside of opam) ocamlbuild 0.14.2 OCamlbuild is a build system with builtin rules to easily build most OCaml projects ocamlfind 1.9.6 A library manager for OCaml pp_loc 2.1.0 Quote and highlight input fragments at a given source location ppx_yojson_conv_lib v0.15.0 Runtime lib for ppx_yojson_conv printbox 0.6.1 Allows to print nested boxes, lists, arrays, tables in several formats printbox-text 0.6.1 Text renderer for printbox, using unicode edges ptime 1.1.0 POSIX time for OCaml re 1.10.4 RE is a regular expression library for OCaml result 1.5 Compatibility Result module seq base Compatibility package for OCaml's standard iterator type starting from 4.07. sqlite3 5.1.0 SQLite3 bindings for OCaml sqlite3_utils 0.4 High-level wrapper around ocaml-sqlite3 stdlib-shims 0.3.0 Backport some of the new stdlib features to older compiler tiny_httpd 0.12 Minimal HTTP server using good old threads topkg 1.0.7 The transitory OCaml software packager uucp 15.0.0 Unicode character properties for OCaml uuidm 0.9.8 Universally unique identifiers (UUIDs) for OCaml uutf 1.0.3 Non-blocking streaming Unicode codec for OCaml yojson 2.0.2 Yojson is an optimized parsing and printing library for the JSON format
The version of Benchpress I used:
<><> Version-specific details <><><><><><><><><><><><><><><><><><><><><><><><><>
version 0.1
repository default
pin git+https://github.com/sneeuwballen/benchpress.git
source-hash 2d84e9d
url.src "git+https://github.com/sneeuwballen/benchpress.git"
homepage "https://github.com/sneeuwballen/benchpress/"
bug-reports "https://github.com/sneeuwballen/benchpress/issues"
dev-repo "git+https://github.com/sneeuwballen/benchpress.git"
authors "Simon Cruanes" "Guillaume Bury"
maintainer "[email protected]"
depends "dune" {>= "2.2"}
"base-unix"
"containers" {>= "3.3" & < "4.0"}
"containers-thread" {>= "3.0" & < "4.0"}
"re" {>= "1.8" & < "2.0"}
"csv"
"cmdliner" {>= "1.1.0"}
"iter" {>= "1.0"}
"logs"
"uuidm"
"base64"
"ptime"
"pp_loc" {>= "2.0" & < "3.0"}
"gnuplot" {>= "0.6" & < "0.8"}
"sqlite3"
"sqlite3_utils" {>= "0.4" & < "0.6"}
"printbox" {>= "0.6"}
"printbox-text" {>= "0.6"}
"ocaml" {>= "4.08"}
synopsis Tool to run one or more logic programs, on a set of files, and collect the results
When looking at a list of results (for instance when filtering problems by name or status), printing the total number of entries that match the filter would be quite useful, ^^
use CCPool
for thread pool, with a kind of:
val map_p : j:int -> ('a -> 'b fut) -> 'a list -> 'b list fut
Particularly for large benchmarks, it would be useful to be able to filter the list of detailed results based on the result/tag of the run.
Once the double wrapping of #39 has been fixed, here are a few feature wish that would be nice to have:
Of course, these two feature wished interact a bit, so ultimately a way for a prover configuration to include which limits the ulimit has to enforce (and removing the wrapper if there are none) would probably be the solution. I'll see if I have the time to work on that in the near future.
edit: in xdg rundir, save a description of the job (a flat json file) and append results as we go. remove file when we're done.
resume
looks for such files and picks up where ti stopped.
Opam pin installation fails if the package irc-client is not previously installed (with an error message indicating to run jbuilder that itself suggests to install the irc-client package).
Should the dependency to irc-client be optional?
par_map
should take a pool as argument, not create onesee https://wwwlehre.dhbw-stuttgart.de/~sschulz/THESES/e_bench2.html for the goals
It looks like the data points given to plots only contain sat/unsat/... results (i.e. builtins tags), but ignores custom tags. For a run that only uses custom tags this result in the following error (from benchpress-server logs):
line 1: warning: Skipping data file with no valid points
gnuplot> plot '-' using 1:2 with linespoints t "dolmen"
^
line 1: x range is invalid
death to toml!
edit: could use decoders-sexp or something like that.
in benchpress serve
, the user should be able to click on run task
, so we need a queue of tasks (also useful for cron tasks, CI, etc.).
simple thing with a level of priority and not much else. run tasks sequentially.
display prover info from this particular DB, which could be custom.
This tool seems very promising! Thanks for making the work available.
I'm trying to get it configured initially. I suspect I'm doing something wrong in the configuration, and I'll keep investigating that, but this behavior seems undesirable regardless, so I figured it merited a report.
When I run benchpress in its currently configured state with default verbosity, I get
$ benchpress run -c config.sexp --task test-on-raft
/bin/sh: 1: z3: not found
/bin/sh: 1: cvc4: not found
testing with 1 provers, 0 problems…
OK
STAT:
provers
----------
sat
----------
unsat
----------
sat+unsat
----------
errors
----------
unknown
----------
timeout
----------
memory
----------
total
----------
total_time
ANALYSIS:
provers
----------
improved
----------
ok
----------
disappoint
----------
bad
----------
errors
----------
total
and a clean exit code. When I pump to max verbosity, I find
[2021-01-26T19:34:15|t0] [info:] run-main.main for paths
[2021-01-26T19:34:15|t0] [debug:api]
create client cl_id=0 host="127.0.0.1" port=8090
[2021-01-26T19:34:15|t0] [debug:]
get-cmd-out "grep -c processor /proc/cpuinfo"
[2021-01-26T19:34:15|t1] [debug:] send progress api to the server
[2021-01-26T19:34:15|t1] [debug:api] make API call task_update
[2021-01-26T19:34:15|t1] [debug:]
error when trying to connect to API: Unix.Unix_error(Unix.ECONNREFUSED, "connect", "")
[2021-01-26T19:34:15|t0] [debug:] par-map: create pool j=3
[2021-01-26T19:34:15|t0] [debug:] par-map: stop pool
testing with 1 provers, 0 problems…
[2021-01-26T19:34:15|t0] [debug:] par-map: create pool j=1
[2021-01-26T19:34:15|t0] [debug:] par-map: stop pool
[2021-01-26T19:34:15|t0] [info:]
benchmark done in 0.072s, uuid=dc58297a-da7a-4fc2-bc4a-02f304d739d6
[2021-01-26T19:34:15|t0] [debug:] saving metadata…
[2021-01-26T19:34:15|t0] [debug:] closing db…
OK
STAT:
provers
----------
sat
----------
unsat
----------
sat+unsat
----------
errors
----------
unknown
----------
timeout
----------
memory
----------
total
----------
total_time
ANALYSIS:
provers
----------
improved
----------
ok
----------
disappoint
----------
bad
----------
errors
----------
total
So it looks like something in my misconfigured system is causing the connection to be refused. But instead of failing with an error, benchproess is reporting everything to be OK.
This is running on an install pinned to master. My system (although not relevant for the silent error):
Linux system76-pc 5.8.0-7630-generic #32~1609193707~20.04~781bb80~dev-Ubuntu SMP Tue Jan 5 21:22:25 U x86_64 x86_64 x86_64 GNU/Linux
logitest-irc
, with its own config perhaps, and launch itprobably need to use query parameters and not just url path in the csv handler…
would be better than "back to root".
Similarly to how provers are identified by a name (rather than their command line), it would be useful to be able to name directories. This would allow a few things such as:
iterate on existing results and use olinq to find problems where the result is different (and time is not too close to timeout); summarize in csv/printbox
if the metadata for a file is already in cache, generate that instead of the lazy loaded version. otherwise do lazy loading.
Probably vega-lite in browser, with some json obtained from the server (we can make a tiny json library for it for the server I think) would be prettier and more convenient than the current generation of gnuplot images.
https://component.kitchen/elix/AutoCompleteComboBox could be nice. Completion should be doable for provers and results (but not problem, too many of them).
On a view such as https://benchpress.cedeela.fr/show_table/res-20200515T222015-2eed2c42-3d64-4611-956f-341d3b2d54ab.sqlite/?offset=0 , it would be useful to be able to list only the results that differ between provers (and this list would be linked to by the numbers in the comparison table of https://benchpress.cedeela.fr/show/res-20200515T222015-2eed2c42-3d64-4611-956f-341d3b2d54ab.sqlite , like is done for each provers' individual results)
-[x] add (proof_produce " --some --additional --args $proof_file")
to prover stanza
(proof_checker "checker $problem --whatever $proof")
to prover stanza(proof_checker <name>)
field in prover)cc @Gbury
attach database $file
to be able to compare 2 distinct runsexternal:file_path
for both stdout and stderrprover-file.out
and prover.file.err
), or a sha1-of-content storageallow, say:
[prover]
cmd = "prover $a $b $file $timeout
a = [a1, a2, a3]
b = [b1, b2]
and test on the whole matrix.
once it's solid
would make it easy to interoperate with other tools, providing json data (and accepting json payloads for things like scheduling tasks).
edit: json, protobuf (ocaml-protoc), … to be decided
I am using Benchpress to track regressions in Alt-Ergo but sometimes I got regressions that I cannot reproduce manually. My guess is that the way Benchpress computes running times is not appropriate for this purpose and the current times can be wrong if the server is overloaded. Thus, I tried to run Benchpress with less jobs to prevent this behavior, but still I have non-replicable regressions.
I could solve my problem by adding the running time without IO operations (the CPU time?) in the database for each tests.
I would be glad to add the feature by myself ;)
useful when it's done by running a reference solver
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.