Slide 1
Slide 1 text
Learn Programming Essence
from Ruby patches
Mitsutaka Mimura
@takkanm
Slide 2
Slide 2 text
Today’s Topic
The aim of the session is to review the
essence of programming, which you
can read from several textbooks for
programming, by reading some Ruby
code as well as libraries.
Slide 3
Slide 3 text
about me
me:
name: Mitsutaka Mimura
github: takkanm
company: EiwaSystemManagement
community: Asakusa.rb
book: PerfectRuby
Slide 4
Slide 4 text
No content
Slide 5
Slide 5 text
No content
Slide 6
Slide 6 text
IUUQCJUMZFTNSVCZLBJHJCFOUP
Slide 7
Slide 7 text
No content
Slide 8
Slide 8 text
l"KZBSJNPDIJTPVOETMJLF"HJMFEBKBSF@DMVCz
.ZCPTTTBJE
Slide 9
Slide 9 text
No content
Slide 10
Slide 10 text
No content
Slide 11
Slide 11 text
Agenda
✤What is a programming knowledge?
✤Learn from patches.
✤Example case
Slide 12
Slide 12 text
What is Programming Knowladge?
Slide 13
Slide 13 text
What is Knowledge of
Programming at Daily
Work?
Slide 14
Slide 14 text
Knowledge of
Programming Used in
Daily Tasks
What is required in daily tasks in knowledge
on the followings.
✤ Language feature
✤ How to use libraries
✤ How to use freameworks.
Slide 15
Slide 15 text
Knowledge of Programming
Seen in Occasions Like
University Leactures
Slide 16
Slide 16 text
Knowledge of Programming
Seen in Occasions Like
University Leactures
✤Algorithm
✤Data Structure
✤Computational Complexity
Slide 17
Slide 17 text
Are they unnecessary at
daily work?
We usually don't have chances to use such
knowledge for writing business codes.
But we believe such textbooks still should hold
much more secrets, such as an algorithm for
writing smarter code or more efficient code.
Slide 18
Slide 18 text
How to study ?
Slide 19
Slide 19 text
Boring Textbooks
From textbooks, we can learn much like
implementation of algorithms or pros/cons of that.
But it is not easy for us to imagine how such
knowledges can be applied to our actual writing
codes.
Slide 20
Slide 20 text
Book ʮPrograming Techniqueʯ
This book is a book that will learn programming
techniques read the source of the UNIX command.
I was able to get some idea of how they have been
utilized in the actual programming is what I
learned at school in this book.
Slide 21
Slide 21 text
Proposal
-FUTUPUIFNPSFGBNJMJBS
UIBUZPVIBWFUPMFBSOUP
SFBEUIFQBUDIPG3VCZ
Slide 22
Slide 22 text
Learn from patches
Slide 23
Slide 23 text
Why patch?
✤ Clear code of interest
✤ Information that the many explanation for
the patch
✤ Choose the ones likely to read to suit your
level
Slide 24
Slide 24 text
What you do learn from a
patch?
✤Understanding of libraries and Ruby
implementation
✤Understanding of the algorithm
✤Practical implementation
Slide 25
Slide 25 text
At an Asakus.rb Meetup
I was taught about Ruby and patches that
improve its performance.
By reading this, I again about improvements due
to change in algorithm.
Slide 26
Slide 26 text
Improving Performance in
Ruby
I might remind you of advanced and esoteric
works like improving GC algorithms or JIT
something like that.
But some patches are not so hard to read.
Slide 27
Slide 27 text
But, You should be learn
about Ruby Interpreter.
That said, you need to know the mechanism of Ruby
even a little to read the patch of Ruby.
So, I will introduce my recommended materials.
Slide 28
Slide 28 text
Ruby Under
a Microscope
Slide 29
Slide 29 text
Ruby Under
a Microscope
Slide 30
Slide 30 text
doc/extention.rdoc (in ruby source tree)
Slide 31
Slide 31 text
yotii23’s
slide :
WALK
AROUND
THE
Ruby
FOREST
MORE
DEEPLY.
Slide 32
Slide 32 text
Example case
Slide 33
Slide 33 text
https://bugs.ruby-lang.org/issues/12142
Slide 34
Slide 34 text
about st_table
Ruby's core holds a 'st_table' structure that have
been used in many codes.
The contents of 'Hash' in Ruby are also contained
within the strcture.
Slide 35
Slide 35 text
review of Hash structure
Slide 36
Slide 36 text
review of Hash
hash = Hash.new
hash[:key_a] = 1
Slide 37
Slide 37 text
review of Hash
)BTI5BCMF
Slide 38
Slide 38 text
review of Hash
,&:
7"-6&
LFZ@B
)BTI5BCMF
Slide 39
Slide 39 text
review of Hash
,&:
7"-6&
LFZ@B
)BTI
GVODUJPO
)BTI5BCMF
Slide 40
Slide 40 text
review for Hash
,&:
7"-6&
LFZ@B
)BTI
GVODUJPO
)BTI5BCMF
Slide 41
Slide 41 text
review for Hash
,&:
7"-6&
LFZ@B
)BTI5BCMF
Slide 42
Slide 42 text
in Ruby
,&:
7"-6&
LFZ@B
Slide 43
Slide 43 text
in Ruby
,&:
7"-6&
LFZ@B
TU@UBCMF
TU@UBCMF@FOUSZ
Slide 44
Slide 44 text
st_table
struct st_table {
const struct st_hash_type *type;
st_index_t num_bins;
unsigned int entries_packed : 1;
st_index_t num_entries : ST_INDEX_BITS - 1;
union {
struct {
struct st_table_entry **bins;
void *private_list_head[2];
} big;
struct {
struct st_packed_entry *entries;
st_index_t real_entries;
} packed;
} as;
};
Slide 45
Slide 45 text
st_table_entry
struct st_table_entry {
st_index_t hash;
st_data_t key;
st_data_t record;
st_table_entry *next;
struct list_node olist;
};
Slide 46
Slide 46 text
Within Ruby
,&:
7"-6&
LFZ@B
TU@UBCMF
TU@UBCMF@FOUSZ
Slide 47
Slide 47 text
Within Ruby
LFZ
SFDPSE
LFZ@B
TU@UBCMFCJOT
TU@UBCMF@FOUSZ
Slide 48
Slide 48 text
Further add a key
LFZ
SFDPSE
LFZ@B
LFZ
SFDPSE
LFZ@C
)BTI
GVODUJPO
Slide 49
Slide 49 text
Within Ruby
LFZ
SFDPSE
LFZ@B
LFZ
SFDPSE
LFZ@C
Slide 50
Slide 50 text
If Hash values conflict
LFZ
SFDPSE
LFZ@B
LFZ
SFDPSE
LFZ@C
LFZ
̍
SFDPSE
LFZ@D
)BTI
GVODUJPO
Slide 51
Slide 51 text
Algorithms to detour
Hash conflict
Slide 52
Slide 52 text
Separate chaining
In case the evaluated values of a hash function are
identical, a single address in a hash table is
allowed to hold multiple elements.
Slide 53
Slide 53 text
Separate chaining
LFZ
SFDPSE
LFZ@B
LFZ
SFDPSE
LFZ@C
LFZ
̍
SFDPSE
LFZ@D
)BTI
GVODUJPO
Slide 54
Slide 54 text
Separate chaining
LFZ
SFDPSE
LFZ@B
LFZ
SFDPSE
LFZ@C
LFZ
̍
SFDPSE
LFZ@D
Slide 55
Slide 55 text
Current Ruby
Ruby use Separate chaining.
Too see its implementation, you can refer
to the function that inserts into hashes.
Slide 56
Slide 56 text
st_insert
int
st_insert(register st_table *table, register st_data_t key, st_data_t
value)
{
st_index_t hash_val;
register st_index_t bin_pos;
register st_table_entry *ptr;
hash_val = do_hash(key, table);
if (table->entries_packed) {
st_index_t i = find_packed_index(table, hash_val, key);
if (i < table->real_entries) {
PVAL_SET(table, i, value);
return 1;
}
add_packed_direct(table, key, value, hash_val);
return 0;
}
Slide 57
Slide 57 text
st_insert
FIND_ENTRY(table, ptr, hash_val, bin_pos);
if (ptr == 0) {
add_direct(table, key, value, hash_val, bin_pos);
return 0;
}
else {
ptr->record = value;
return 1;
}
}
Slide 58
Slide 58 text
add_direct
static inline void
add_direct(st_table *table, st_data_t key, st_data_t value,
st_index_t hash_val, register st_index_t bin_pos)
{
register st_table_entry *entry;
if (table->num_entries > ST_DEFAULT_MAX_DENSITY * table->num_bins) {
rehash(table);
bin_pos = hash_pos(hash_val, table->num_bins);
}
entry = new_entry(table, key, value, hash_val, bin_pos);
list_add_tail(st_head(table), &entry->olist);
table->num_entries++;
}
Slide 59
Slide 59 text
new_entry
static inline st_table_entry *
new_entry(st_table * table, st_data_t key, st_data_t value,
st_index_t hash_val, register st_index_t bin_pos)
{
register st_table_entry *entry = st_alloc_entry();
entry->next = table->bins[bin_pos];
table->bins[bin_pos] = entry;
entry->hash = hash_val;
entry->key = key;
entry->record = value;
return entry;
}
Slide 60
Slide 60 text
If conflict Hash value
LFZ
SFDPSE
LFZ@C
LFZ
̍
SFDPSE
LFZ@D
)BTI
GVODUJPO
Slide 61
Slide 61 text
If conflict Hash value
LFZ
SFDPSE
LFZ@C
LFZ
̍
SFDPSE
LFZ@D
Slide 62
Slide 62 text
If conflict Hash value
LFZ
SFDPSE
LFZ@C
LFZ
̍
SFDPSE
LFZ@D
Slide 63
Slide 63 text
Feature #12142
Hash tables with open addressing
switching to open addressing hash tables
for access by keys.
Removing hash collision lists lets us avoid
*pointer chasing*, a common problem that
produces bad data locality. I see a tendency
to move from chaining hash tables to open
addressing hash tables due to their better fit
to modern CPU memory organizations.
Slide 64
Slide 64 text
Open addressing
A way to insert all Hash elements in a
Hash table. When the evaluation results of
Hash function conflict, the element is
placed in another empty space.
Slide 65
Slide 65 text
If conflict Hash value
LFZ
SFDPSE
LFZ@C
LFZ
̍
SFDPSE
LFZ@D
)BTI
GVODUJPO
Slide 66
Slide 66 text
If conflict Hash value
LFZ
SFDPSE
LFZ@C
LFZ
̍
SFDPSE
LFZ@D
)BTI
GVODUJPO
JOD
Slide 67
Slide 67 text
If conflict Hash value
LFZ
SFDPSE
LFZ@C
LFZ
̍
SFDPSE
LFZ@D
Slide 68
Slide 68 text
Read patch
Slide 69
Slide 69 text
Corresponding Patches
Half an year has passed since the original
proposal, and various refinements have
been adopted since. This talk will be based
on the latest attached st_table_with_array2
branch of funny-falcon/ruby.
Slide 70
Slide 70 text
st_table
struct st_table {
const struct st_hash_type *type;
union {
struct st_table_entry* entries;
st_idx_t* bins;
uint8_t* smallbins;
uint16_t* medbins;
} as;
st_idx_t num_entries;
st_idx_t first, last;
unsigned sz : 8;
unsigned rebuild_num : 24;
};
Slide 71
Slide 71 text
st_table_entry
struct st_table_entry {
st_idx_t hash;
st_idx_t next;
st_data_t key;
st_data_t record;
};
Slide 72
Slide 72 text
new st_table
TU@UBCMF
CJOT FOUSJFT
Slide 73
Slide 73 text
st_insert
int
st_insert(register st_table *table, register st_data_t key, st_data_t value)
{
st_idx_t hash_val, idx;
hash_val = do_hash(key, table);
idx = find_entry(table, key, hash_val);
if (idx == IDX_NULL) {
add_direct(table, key, value, hash_val);
return 0;
}
else {
table->as.entries[idx].record = value;
return 1;
}
}
Slide 74
Slide 74 text
find_entry
static inline st_idx_t
find_entry(const st_table *table, st_data_t key, st_idx_t hash_val)
{
unsigned rebuild_num ST_UNUSED = table->rebuild_num;
if (st_sz[table->sz].nbins == 0) {
st_idx_t idx = table->first, last = table->last;
st_table_entry* ptr = &table->as.entries[idx];
for (; idx < last; idx++, ptr++) {
if (ptr->hash == hash_val && EQUAL(table, key, ptr)) {
return idx;
}
}
st_assert(rebuild_num == table->rebuild_num);
return IDX_NULL;
Slide 75
Slide 75 text
find_entry
} else {
st_idx_t bin_pos = hash_pos(hash_val, table->sz);
st_idx_t idx = bin_get(table, bin_pos);
FOUND_ENTRY;
while (PTR_NOT_EQUAL(table, idx, hash_val, key)) {
COLLISION;
idx = table->as.entries[idx].next;
}
st_assert(rebuild_num == table->rebuild_num);
return idx;
}
}
Slide 76
Slide 76 text
add_direct
static inline void
add_direct(st_table *table, st_data_t key, st_data_t value,
st_idx_t hash_val)
{
register st_table_entry *entry;
st_idx_t en_idx, bin_pos;
if (table->last == st_sz[table->sz].nentries) {
st_rehash(table);
}
Slide 77
Slide 77 text
add_direct
en_idx = table->last;
table->last++;
entry = &table->as.entries[en_idx];
if (st_sz[table->sz].nbins != 0) {
bin_pos = hash_pos(hash_val, table->sz);
entry->next = bin_get(table, bin_pos);
bin_set(table, bin_pos, en_idx);
}
entry->hash = hash_val;
entry->key = key;
entry->record = value;
table->num_entries++;
}
Slide 78
Slide 78 text
new st_table
TU@UBCMF
CJOT FOUSJFT LFZ WBMVF IBTI
LFZ@B
Slide 79
Slide 79 text
TU@UBCMF
CJOT FOUSJFT LFZ WBMVF IBTI
LFZ@B
new st_table
LFZ@B
Slide 80
Slide 80 text
̌
TU@UBCMF
CJOT FOUSJFT
LFZ@B
LFZ WBMVF IBTI
LFZ@B
new st_table
Slide 81
Slide 81 text
̌
TU@UBCMF
CJOT FOUSJFT
LFZ@B
LFZ WBMVF IBTI
LFZ@C
new st_table
Slide 82
Slide 82 text
̌
TU@UBCMF
CJOT FOUSJFT
LFZ@B
LFZ@C
LFZ WBMVF IBTI
LFZ@C
new st_table
Slide 83
Slide 83 text
̌
̍
TU@UBCMF
CJOT FOUSJFT
LFZ@B
LFZ@C
LFZ WBMVF IBTI
LFZ@C
new st_table
Slide 84
Slide 84 text
̌
̍
TU@UBCMF
CJOT FOUSJFT LFZ WBMVF IBTI
LFZ@D
new st_table
LFZ@B
LFZ@C
Slide 85
Slide 85 text
̌
̍
TU@UBCMF
CJOT FOUSJFT LFZ WBMVF IBTI
LFZ@D
LFZ@D
new st_table
LFZ@B
LFZ@C
Slide 86
Slide 86 text
̌
̍
TU@UBCMF
CJOT FOUSJFT LFZ WBMVF IBTI
LFZ@D
̍
new st_table
LFZ@D
LFZ@B
LFZ@C
Slide 87
Slide 87 text
̌
̎
TU@UBCMF
CJOT FOUSJFT LFZ WBMVF IBTI
LFZ@D
new st_table
̍
LFZ@D
LFZ@B
LFZ@C
Slide 88
Slide 88 text
Feeling after read patch
✤ By reading the patch of the description, it
is easy to guess the contents, it became
help Read.
✤ Patch format was easy to follow the
changes.
✤ Than what you have learned in the book,
knowing the practical implementation.
Slide 89
Slide 89 text
Conclusion
Slide 90
Slide 90 text
Conclusion
By reading real code, We suppose you will be able
to tell the scenarios where ideas described in
books are used. By reading patches, I think you can
easily tell the main changes you are interested in.
I recommend you to try code-reading to rediscover
the knowledge of programming you learned. How
about that?