Replace S3 dispatch by C implementation #400

mgirlich · 2023-08-23T07:48:46Z

Closes #386.

library(xml2)

data <- read_xml("http://aiweb.cs.washington.edu/research/projects/xmltk/xmldata/data/courses/reed.xml")
many_elts <- data |> 
  xml_contents() |> 
  xml_contents() |>
  xml_contents()

path <- xml2_example("cd_catalog.xml")
xml <- read_xml(path)
cds <- xml |> xml_children()
cd_contents <- cds |> xml_contents()

bench::mark(
  cds = xml_name(cds),
  cd_contents = xml_name(cd_contents),
  many_elts = xml_name(many_elts),
  check = FALSE
)

# DEV
#> # A tibble: 3 × 6
#>   expression       min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>  <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 cds           78.6µs   82.5µs   11139.     6.95KB     23.7
#> 2 cd_contents  447.2µs  480.9µs    1940.     1.27KB     21.2
#> 3 many_elts     28.2ms   30.1ms      33.3   65.72KB     42.8

# This PR
#> # A tibble: 3 × 6
#>   expression       min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>  <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 cds           3.52µs   3.87µs   214805.    2.66KB    21.5 
#> 2 cd_contents  13.03µs   14.1µs    61534.    1.27KB    18.5 
#> 3 many_elts     1.23ms   1.71ms      572.   65.72KB     8.36

^{Created on 2023-08-23 with reprex v2.0.2}

More benchmarks

library(xml2)

data <- read_xml("http://aiweb.cs.washington.edu/research/projects/xmltk/xmldata/data/courses/reed.xml")
many_elts <- data |> 
  xml_contents() |> 
  xml_contents() |>
  xml_contents()

path <- xml2_example("cd_catalog.xml")
xml <- read_xml(path)
cds <- xml |> xml_children()
cd_contents <- cds |> xml_contents()

bench::mark(
  cds = xml_name(cds),
  cd_contents = xml_name(cd_contents),
  many_elts = xml_name(many_elts),
  check = FALSE
)
#> # A tibble: 3 × 6
#>   expression       min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>  <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 cds             66µs   78.6µs   11863.     6.95KB     23.8
#> 2 cd_contents  378.9µs  449.6µs    2069.     1.27KB     23.5
#> 3 many_elts     24.3ms   27.9ms      36.3   65.72KB     40.8

#> # A tibble: 3 × 6
#>   expression       min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>  <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 cds           2.85µs   3.07µs   277778.    2.66KB     0   
#> 2 cd_contents  13.17µs  14.23µs    62456.    1.27KB    18.7 
#> 3 many_elts      1.1ms   1.63ms      616.   65.72KB     8.29

bench::mark(
  cds = xml_attr(cds, "test"),
  cd_contents = xml_attr(cd_contents, "test"),
  many_elts = xml_attr(many_elts, "test"),
  check = FALSE
)

#> # A tibble: 3 × 6
#>   expression       min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>  <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 cds           80.7µs   95.9µs    9750.     8.66KB     25.5
#> 2 cd_contents  464.3µs  538.8µs    1779.     1.27KB     23.6
#> 3 many_elts     29.2ms   30.4ms      32.9   65.72KB    132.

#> # A tibble: 3 × 6
#>   expression       min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>  <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 cds           2.77µs   3.01µs   292658.    3.44KB     29.3
#> 2 cd_contents   8.88µs   9.66µs    87308.    1.27KB     26.2
#> 3 many_elts   814.71µs   1.02ms      889.   65.72KB     10.4

bench::mark(
  cds = xml_attrs(cds),
  cd_contents = xml_attrs(cd_contents),
  many_elts = xml_attrs(many_elts),
  check = FALSE
)

#> # A tibble: 3 × 6
#>   expression       min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>  <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 cds           71.6µs   83.8µs   11029.     6.48KB     25.6
#> 2 cd_contents  401.2µs  475.8µs    1997.     1.27KB     26.8
#> 3 many_elts     24.5ms   25.9ms      39.1   65.72KB    102.

#> # A tibble: 3 × 6
#>   expression       min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>  <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 cds           3.54µs   4.39µs   194171.    3.02KB     38.8
#> 2 cd_contents  15.28µs  18.69µs    44680.    1.27KB     44.7
#> 3 many_elts     1.07ms   1.42ms      671.   65.72KB     29.9

bench::mark(
  cds = xml_text(cds),
  cd_contents = xml_text(cd_contents),
  many_elts = xml_text(many_elts),
  check = FALSE
)

#> # A tibble: 3 × 6
#>   expression       min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>  <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 cds           94.8µs  110.1µs    7962.    11.24KB     22.1
#> 2 cd_contents  526.7µs  653.4µs    1241.     1.27KB     19.2
#> 3 many_elts     43.7ms   48.6ms      21.3   65.72KB     56.7

#> # A tibble: 3 × 6
#>   expression       min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>  <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 cds           11.8µs  12.22µs    74143.    5.14KB     7.42
#> 2 cd_contents  37.14µs  42.75µs    22727.    1.27KB     9.09
#> 3 many_elts     1.94ms   2.36ms      418.   65.72KB     6.30

bench::mark(
  cds = xml_length(cds),
  cd_contents = xml_length(cd_contents),
  many_elts = xml_length(many_elts),
  check = FALSE
)

#> # A tibble: 3 × 6
#>   expression       min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>  <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 cds           75.1µs   85.1µs    8996.     8.69KB     18.9
#> 2 cd_contents  417.2µs  475.9µs    1739.       672B     21.2
#> 3 many_elts     23.6ms   27.2ms      35.7   32.88KB     45.9

#> # A tibble: 3 × 6
#>   expression       min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>  <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 cds            1.9µs   2.06µs   418749.    2.41KB      0  
#> 2 cd_contents   6.13µs   6.71µs   121050.      672B     36.3
#> 3 many_elts   712.37µs 852.89µs     1075.   32.88KB     14.7

bench::mark(
  cds = xml_path(cds),
  cd_contents = xml_path(cd_contents),
  many_elts = xml_path(many_elts),
  check = FALSE
)

#> # A tibble: 3 × 6
#>   expression       min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>  <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 cds           80.9µs   96.3µs    8871.     6.12KB     18.9
#> 2 cd_contents  499.1µs  614.8µs    1412.     1.27KB     14.6
#> 3 many_elts     56.2ms     67ms      15.6   65.72KB     26.0

#> # A tibble: 3 × 6
#>   expression       min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>  <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 cds           15.7µs   18.5µs   52116.     2.46KB     5.21
#> 2 cd_contents  117.7µs  134.9µs    7305.     1.27KB     2.01
#> 3 many_elts     25.6ms     29ms      33.5   65.72KB     0

bench::mark(
  cds = xml_type(cds),
  cd_contents = xml_type(cd_contents),
  many_elts = xml_type(many_elts),
  check = FALSE
)

#> # A tibble: 3 × 6
#>   expression       min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>  <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 cds           45.1µs   53.1µs   15496.      8.1KB     21.1
#> 2 cd_contents  232.4µs  281.6µs    2984.     1.92KB     21.3
#> 3 many_elts     16.1ms   17.8ms      52.9    98.6KB     26.5

#> # A tibble: 3 × 6
#>   expression       min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>  <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 cds           1.85µs   2.22µs   360872.    5.73KB     36.1
#> 2 cd_contents   6.04µs    8.2µs   104980.    1.92KB     31.5
#> 3 many_elts   702.33µs    870µs     1076.    98.6KB     15.0

^{Created on 2023-08-30 with reprex v2.0.2}

mgirlich · 2023-08-23T09:41:35Z

@hadley It would be great to get your opinion of whether this approach is fine for you. It kind of works against the dispatch system. But I'm also not quite sure whether the dispatch is really needed as there are only three classes (<xml_missing>, <xml_node> and <xml_nodeset>) and I don't think the user really needs to extend them.

The performance gains are quite big (around a factor of 20), which would be nice for e.g. the {paws} package (a package to work with AWS).

Overall, the code is pretty redundant so we could also use a macro or some other CPP techniques but I'm not so familiar with C/CPP to know about the potential downsides.

hadley

I think switching from S3 dispatch to something custom at the C-level is fine, although as I noted I think providing a helper that makes it possible to use switch will make the implementation cleaner (and a little easier to extend in the unlikely event that is necessary).

I'm not too worried about the duplication, although the idiosyncratic mix of C and C++ is going to make maintenance generally harder. Possibly might be better to use cpp11 and some of the vector helpers that it provides.

R/xml_text.R

hadley · 2023-08-23T20:11:14Z

tests/testthat/test-xml_name.R

@@ -1,3 +1,13 @@
+test_that("xml_name() returns the name", {
+  x <- sample_nodeset()


IMO it would be better to copy and paste the contents of sample_nodeset() here; otherwise it's hard to verify that the test is correct.

src/xml2_node.cpp

inst/include/xml2_types.h

DavisVaughan · 2023-09-01T19:20:24Z

src/xml2_node.cpp

+    XPtrNode node(node_sxp);
+
+    std::string name = nodeName(node.checked_get(), nsMap);
+    out = Rf_mkCharLenCE(name.c_str(), name.size(), CE_UTF8);


I suppose this has always been the case, but if we have an R error anywhere in here, then it will skip over any C++ destructors (like for XPtrNode, if it has one, and std::string) on the way out

I don't think there is a reason to try and improve this in this PR, but I will keep an eye out for potentially problematic cases as I continue my review

Yeah, we have an odd mix of C and C++ in this package which we should try and fix at some point.

src/xml2_node.cpp

DavisVaughan · 2023-09-01T19:38:37Z

src/xml2_node.cpp

+    if (n == 0) {
+      return Rf_ScalarInteger(0);
+    }


I think everything is fine if you remove this. Generally i try and reserve early exits for cases where they are absolutely necessary (like if your algorithm requires >=2 elements or something, like most sorting algos)

Oh wait, I see what you are doing here. This seems odd. If the nodeset is length 0, shouldn't the output be length 0? i.e. integer()? In which case you could definitely just remove this then it would do the right thing.

Maybe this is some old behavior I don't know about

I also noticed this and found it quite weird which is why I opened an issue #404. I wanted to tackle that in a separate PR, forgot to mention this here.

tests/testthat/test-xml_type.R

hadley · 2023-11-30T22:11:27Z

This breaks fhircracker

setClass(
	Class = "fhir_bundle_xml",
	contains = c("fhir_resource_xml", "fhir_bundle"),
	slots = c(next_link = "fhir_url", self_link = "fhir_url"),
	prototype = prototype(xml2::read_xml(x = "<Bundle></Bundle>"))
)

hadley · 2023-11-30T22:15:24Z

I'm guessing this is because while inherits(x, "xml_node") is still true, Rf_inherits() doesn't handle S4 classes.

Yeah — looks like it https://github.com/wch/r-source/blob/d6dad605b05810cb43f991d292169a2cd436a818/src/include/Rinlinedfuns.h#L774.

mgirlich marked this pull request as draft August 23, 2023 07:49

Implement xml_name.xml_nodeset() in C

8f19086

mgirlich force-pushed the loop-in-c branch from d2db44c to 8f19086 Compare August 23, 2023 08:49

mgirlich added 3 commits August 23, 2023 08:59

Implement xml_text.xml_nodeset() in C

073a0a5

Use xml_text.xml_nodeset() in xml_double/integer()

3685c8e

Implement xml_type.xml_nodeset() in C

5784f67

hadley approved these changes Aug 23, 2023

View reviewed changes

mgirlich added 6 commits August 29, 2023 08:42

Remove S3 dispatch in xml_name()

34a5f12

Remove S3 dispatch in xml_text()

653e56e

Simplify xml_double() and xml_integer()

2aa3ca1

Remove S3 dispatch in xml_type()

84674e1

Inline sample_nodeset() in tests

69da196

Move NodeType enum to xml2_types.h

03445ec

mgirlich changed the title ~~Implement xml_name.xml_nodeset() in C~~ Implement xml_name(), xml_text() and xml_type() in C Aug 29, 2023

mgirlich force-pushed the loop-in-c branch from c215e94 to 03445ec Compare August 29, 2023 13:01

mgirlich added 7 commits August 29, 2023 13:15

Document xml_integer()

2293448

Implement xml_length() in C

4a92b3b

Implement xml_attr() in C

1406af6

Implement xml_attrs() in C

3e2049e

Implement xml_path() in C

08ba881

Fix return style

efc24cf

Fix compilation warning

6a5624f

mgirlich changed the title ~~Implement xml_name(), xml_text() and xml_type() in C~~ Replace S3 dispatch by C implementation Aug 29, 2023

mgirlich marked this pull request as ready for review August 30, 2023 06:27

DavisVaughan approved these changes Sep 1, 2023

View reviewed changes

mgirlich added 4 commits September 4, 2023 07:05

Move getNodeType() to xml2_utils.h

7c1f008

Don't align : in switches

10fcf60

Pass string by reference

4368951

Use array indexing for pointer

a1ba915

mgirlich added 3 commits September 4, 2023 07:14

Fix test description

71b2160

Add stop_unexpected_node_type()

5d3071f

Fix compiler warning

fe5f269

mgirlich mentioned this pull request Sep 4, 2023

Use cpp11 #405

Open

hadley merged commit 1c4453f into r-lib:main Nov 2, 2023
12 checks passed

hadley added a commit that referenced this pull request Nov 30, 2023

Add news bullet for #400

92190df

hadley mentioned this pull request Nov 30, 2023

Handle inheritance for S4 objects #425

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Replace S3 dispatch by C implementation #400

Replace S3 dispatch by C implementation #400

mgirlich commented Aug 23, 2023 •

edited

Loading

mgirlich commented Aug 23, 2023

hadley left a comment

hadley Aug 23, 2023

DavisVaughan Sep 1, 2023

hadley Sep 1, 2023

DavisVaughan Sep 1, 2023

DavisVaughan Sep 1, 2023

mgirlich Sep 4, 2023

hadley commented Nov 30, 2023

hadley commented Nov 30, 2023 •

edited

Loading

		@@ -1,3 +1,13 @@
		test_that("xml_name() returns the name", {
		x <- sample_nodeset()

Replace S3 dispatch by C implementation #400

Replace S3 dispatch by C implementation #400

Conversation

mgirlich commented Aug 23, 2023 • edited Loading

mgirlich commented Aug 23, 2023

hadley left a comment

Choose a reason for hiding this comment

hadley Aug 23, 2023

Choose a reason for hiding this comment

DavisVaughan Sep 1, 2023

Choose a reason for hiding this comment

hadley Sep 1, 2023

Choose a reason for hiding this comment

DavisVaughan Sep 1, 2023

Choose a reason for hiding this comment

DavisVaughan Sep 1, 2023

Choose a reason for hiding this comment

mgirlich Sep 4, 2023

Choose a reason for hiding this comment

hadley commented Nov 30, 2023

hadley commented Nov 30, 2023 • edited Loading

mgirlich commented Aug 23, 2023 •

edited

Loading

hadley commented Nov 30, 2023 •

edited

Loading