High level conclusions

Fuzzer fuzz_normalizer is blocked:

The runtime code coverage of fuzz_normalizer covers 28.57% of its statically reachable code. This means there is some place that blocks the fuzzer to continue exploring more code at run time.

Fuzzer fuzz_split_xml is blocked:

The runtime code coverage of fuzz_split_xml covers 7.692% of its statically reachable code. This means there is some place that blocks the fuzzer to continue exploring more code at run time.

Fuzzer fuzz_detokenize is blocked:

The runtime code coverage of fuzz_detokenize covers 13.63% of its statically reachable code. This means there is some place that blocks the fuzzer to continue exploring more code at run time.

Fuzzer fuzz_tokenizer is blocked:

The runtime code coverage of fuzz_tokenizer covers 25.80% of its statically reachable code. This means there is some place that blocks the fuzzer to continue exploring more code at run time.

Reachability and coverage overview

Functions statically reachable by fuzzers

32 / 71

Cyclomatic complexity statically reachable by fuzzers

107 / 243

Runtime code coverage of functions

24 / 71

The following table shows data about each function in the project. The functions included in this table correspond to all functions that exist in the executables of the fuzzers. As such, there may be functions that are from third-party libraries.

For further technical details on the meaning of columns in the below table, please see the Glossary .

Func name	Functions filename	Args	Function call depth	Reached by Fuzzers	Runtime reached by Fuzzers	Combined reached by Fuzzers	Fuzzers runtime hit	Func lines hit %	I Count	BB Count	Cyclomatic complexity	Functions reached	Reached by functions	Accumulated cyclomatic complexity	Undiscovered complexity

Fuzzer: fuzz_tokenizer

Call tree

The calltree shows the control flow of the fuzzer. This is overlaid with coverage information to display how much of the potential code a fuzzer can reach is in fact covered at runtime. In the following there is a link to a detailed calltree visualisation as well as a bitmap showing a high-level view of the calltree. For further information about these topics please see the glossary for full calltree and calltree overview

Call tree overview bitmap:

The distribution of callsites in terms of coloring is

Color	Runtime hitcount	Callsite count	Percentage
red	0	14	25.4%
gold	[1:9]	0	0.0%
yellow	[10:29]	0	0.0%
greenyellow	[30:49]	0	0.0%
lawngreen	50+	41	74.5%
All colors	55	100

Fuzz blockers

The following nodes represent call sites where fuzz blockers occur.

Amount of callsites blocked	Calltree index	Parent function	Callsite	Largest blocked function
5	15	sacremoses.corpus.NonbreakingPrefixes.words	call site: 00015	sacremoses.tokenize.MosesTokenizer.has_numeric_only
3	7	sacremoses.tokenize.MosesTokenizer.__init__	call site: 00007	sacremoses.corpus.NonbreakingPrefixes.words
2	3	...fuzz_tokenizer.TestOneInput	call site: 00003	fdp.ConsumeIntInRange
1	35	sacremoses.tokenize.MosesTokenizer.__init__	call site: 00035	sacremoses.tokenize.MosesTokenizer.penn_tokenize
1	44	sacremoses.tokenize.MosesTokenizer.handles_nonbreaking_prefixes	call site: 00044	sacremoses.tokenize.MosesTokenizer.isanyalpha
1	48	sacremoses.tokenize.MosesTokenizer.isanyalpha	call site: 00048	sacremoses.tokenize.MosesTokenizer.islower
1	51	sacremoses.tokenize.MosesTokenizer.islower	call site: 00051	re.search

Runtime coverage analysis

Covered functions

35

Functions that are reachable but not covered

23

Reachable functions

31

Percentage of reachable functions covered

25.81%

NB: The sum of covered functions and functions that are reachable but not covered need not be equal to Reachable functions . This is because the reachability analysis is an approximation and thus at runtime some functions may be covered that are not included in the reachability analysis. This is a limitation of our static analysis capabilities.

Warning: The number of covered functions are larger than the number of reachable functions. This means that there are more functions covered at runtime than are extracted using static analysis. This is likely a result of the static analysis component failing to extract the right call graph or the coverage runtime being compiled with sanitizers in code that the static analysis has not analysed. This can happen if lto/gold is not used in all places that coverage instrumentation is used.

Function name	source code lines	source lines hit	percentage hit

Files reached

filename	functions hit
/	1
...fuzz_tokenizer	5
sacremoses.tokenize	22
sacremoses.corpus	6

Fuzzer: fuzz_detokenize

Call tree

The calltree shows the control flow of the fuzzer. This is overlaid with coverage information to display how much of the potential code a fuzzer can reach is in fact covered at runtime. In the following there is a link to a detailed calltree visualisation as well as a bitmap showing a high-level view of the calltree. For further information about these topics please see the glossary for full calltree and calltree overview

Call tree overview bitmap:

The distribution of callsites in terms of coloring is

Color	Runtime hitcount	Callsite count	Percentage
red	0	78	72.2%
gold	[1:9]	0	0.0%
yellow	[10:29]	0	0.0%
greenyellow	[30:49]	0	0.0%
lawngreen	50+	30	27.7%
All colors	108	100

Fuzz blockers

The following nodes represent call sites where fuzz blockers occur.

Amount of callsites blocked	Calltree index	Parent function	Callsite	Largest blocked function
58	49	sacremoses.tokenize.MosesTokenizer.__init__	call site: 00049	sacremoses.cli.truecase_file
10	10	sacremoses.normalize.MosesPunctNormalizer.__init__	call site: 00010	sacremoses.cli.tokenize_file
5	29	sacremoses.corpus.NonbreakingPrefixes.words	call site: 00029	sacremoses.tokenize.MosesTokenizer.has_numeric_only
3	21	sacremoses.tokenize.MosesTokenizer.__init__	call site: 00021	sacremoses.corpus.NonbreakingPrefixes.words
2	6	...fuzz_detokenize.TestOneInput	call site: 00006	sacremoses.cli.normalize_file

Runtime coverage analysis

Covered functions

35

Functions that are reachable but not covered

57

Reachable functions

66

Percentage of reachable functions covered

13.64%

NB: The sum of covered functions and functions that are reachable but not covered need not be equal to Reachable functions . This is because the reachability analysis is an approximation and thus at runtime some functions may be covered that are not included in the reachability analysis. This is a limitation of our static analysis capabilities.

Function name	source code lines	source lines hit	percentage hit

Files reached

filename	functions hit
/	1
...fuzz_detokenize	5
sacremoses.tokenize	12
sacremoses.cli	23
sacremoses.normalize	2
sacremoses.util	6
sacremoses.corpus	6
sacremoses.truecase	22

Fuzzer: fuzz_split_xml

Call tree

The calltree shows the control flow of the fuzzer. This is overlaid with coverage information to display how much of the potential code a fuzzer can reach is in fact covered at runtime. In the following there is a link to a detailed calltree visualisation as well as a bitmap showing a high-level view of the calltree. For further information about these topics please see the glossary for full calltree and calltree overview

Call tree overview bitmap:

The distribution of callsites in terms of coloring is

Color	Runtime hitcount	Callsite count	Percentage
red	0	4	17.3%
gold	[1:9]	0	0.0%
yellow	[10:29]	0	0.0%
greenyellow	[30:49]	0	0.0%
lawngreen	50+	19	82.6%
All colors	23	100

Fuzz blockers

The following nodes represent call sites where fuzz blockers occur.

Amount of callsites blocked	Calltree index	Parent function	Callsite	Largest blocked function
3	9	sacremoses.truecase.MosesTruecaser.split_xml	call site: 00009	re.search
1	20	sacremoses.truecase.MosesTruecaser.split_xml	call site: 00020	xml_cognates.group

Runtime coverage analysis

Covered functions

35

Functions that are reachable but not covered

12

Reachable functions

13

Percentage of reachable functions covered

7.69%

NB: The sum of covered functions and functions that are reachable but not covered need not be equal to Reachable functions . This is because the reachability analysis is an approximation and thus at runtime some functions may be covered that are not included in the reachability analysis. This is a limitation of our static analysis capabilities.

Warning: The number of covered functions are larger than the number of reachable functions. This means that there are more functions covered at runtime than are extracted using static analysis. This is likely a result of the static analysis component failing to extract the right call graph or the coverage runtime being compiled with sanitizers in code that the static analysis has not analysed. This can happen if lto/gold is not used in all places that coverage instrumentation is used.

Function name	source code lines	source lines hit	percentage hit

Files reached

filename	functions hit
/	1
...fuzz_split_xml	4
sacremoses.truecase	8

Fuzzer: fuzz_normalizer

Call tree

The calltree shows the control flow of the fuzzer. This is overlaid with coverage information to display how much of the potential code a fuzzer can reach is in fact covered at runtime. In the following there is a link to a detailed calltree visualisation as well as a bitmap showing a high-level view of the calltree. For further information about these topics please see the glossary for full calltree and calltree overview

Call tree overview bitmap:

The distribution of callsites in terms of coloring is

Color	Runtime hitcount	Callsite count	Percentage
red	0	8	44.4%
gold	[1:9]	0	0.0%
yellow	[10:29]	0	0.0%
greenyellow	[30:49]	0	0.0%
lawngreen	50+	10	55.5%
All colors	18	100

Fuzz blockers

The following nodes represent call sites where fuzz blockers occur.

Amount of callsites blocked	Calltree index	Parent function	Callsite	Largest blocked function
4	8	sacremoses.normalize.MosesPunctNormalizer.__init__	call site: 00008	sacremoses.normalize.MosesPunctNormalizer.normalize
2	3	...fuzz_normalizer.TestOneInput	call site: 00003	fdp.ConsumeIntInRange
2	14	sacremoses.normalize.MosesPunctNormalizer.normalize	call site: 00014	sacremoses.normalize.MosesPunctNormalizer.remove_control_chars

Runtime coverage analysis

Covered functions

35

Functions that are reachable but not covered

10

Reachable functions

14

Percentage of reachable functions covered

28.57%

NB: The sum of covered functions and functions that are reachable but not covered need not be equal to Reachable functions . This is because the reachability analysis is an approximation and thus at runtime some functions may be covered that are not included in the reachability analysis. This is a limitation of our static analysis capabilities.

Warning: The number of covered functions are larger than the number of reachable functions. This means that there are more functions covered at runtime than are extracted using static analysis. This is likely a result of the static analysis component failing to extract the right call graph or the coverage runtime being compiled with sanitizers in code that the static analysis has not analysed. This can happen if lto/gold is not used in all places that coverage instrumentation is used.

Function name	source code lines	source lines hit	percentage hit

Files reached

filename	functions hit
/	1
...fuzz_normalizer	5
sacremoses.normalize	8

Optimal target analysis

Remaining optimal interesting functions

The following table shows a list of functions that are optimal targets. Optimal targets are identified by finding the functions that in combination, yield a high code coverage.

Func name	Functions filename	Arg count	Args	Function depth	hitcount	instr count	bb count	cyclomatic complexity	Reachable functions	Incoming references	total cyclomatic complexity	Unreached complexity

Implementing fuzzers that target the above functions will improve reachability such that it becomes:

Functions statically reachable by fuzzers

32 / 71

Cyclomatic complexity statically reachable by fuzzers

107 / 243

All functions overview

If you implement fuzzers for these functions, the status of all functions in the project will be:

Func name	Functions filename	Args	Function call depth	Reached by Fuzzers	Runtime reached by Fuzzers	Combined reached by Fuzzers	Fuzzers runtime hit	Func lines hit %	I Count	BB Count	Cyclomatic complexity	Functions reached	Reached by functions	Accumulated cyclomatic complexity	Undiscovered complexity

This sections provides heuristics that can be used as input to a fuzz engine when running a given fuzz target. The current focus is on providing input that is usable by libFuzzer.

/src/fuzz_tokenizer.py

Dictionary

Use this with the libFuzzer -dict=DICT.file flag

Fuzzer function priority

Use one of these functions as input to libfuzzer with flag: -focus_function name

-focus_function=['sacremoses.corpus.NonbreakingPrefixes.words', 'sacremoses.tokenize.MosesTokenizer.__init__', '...fuzz_tokenizer.TestOneInput', 'sacremoses.tokenize.MosesTokenizer.handles_nonbreaking_prefixes', 'sacremoses.tokenize.MosesTokenizer.isanyalpha', 'sacremoses.tokenize.MosesTokenizer.islower']

/src/fuzz_detokenize.py

Dictionary

Use this with the libFuzzer -dict=DICT.file flag

Fuzzer function priority

Use one of these functions as input to libfuzzer with flag: -focus_function name

-focus_function=['sacremoses.tokenize.MosesTokenizer.__init__', 'sacremoses.normalize.MosesPunctNormalizer.__init__', 'sacremoses.corpus.NonbreakingPrefixes.words', '...fuzz_detokenize.TestOneInput']

/src/fuzz_split_xml.py

Dictionary

Use this with the libFuzzer -dict=DICT.file flag

Fuzzer function priority

Use one of these functions as input to libfuzzer with flag: -focus_function name

-focus_function=['sacremoses.truecase.MosesTruecaser.split_xml']

/src/fuzz_normalizer.py

Dictionary

Use this with the libFuzzer -dict=DICT.file flag

Fuzzer function priority

Use one of these functions as input to libfuzzer with flag: -focus_function name

-focus_function=['sacremoses.normalize.MosesPunctNormalizer.__init__', '...fuzz_normalizer.TestOneInput', 'sacremoses.normalize.MosesPunctNormalizer.normalize']

This section shows analysis of runtime coverage data.

For futher technical details on how this section is generated, please see the Glossary .

Complex functions with low coverage

Func name	Function total lines	Lines covered at runtime	percentage covered	Reached by fuzzers
sacremoses.tokenize.MosesTokenizer.tokenize	62	23	37.09%	['fuzz_detokenize', 'fuzz_split_xml', 'fuzz_normalizer', 'fuzz_tokenizer']
re.compile	2790	720	25.80%	['fuzz_detokenize', 'fuzz_tokenizer']
text.strip	173	50	28.90%	['fuzz_split_xml', 'fuzz_normalizer', 'fuzz_tokenizer']
regex.sub	268	117	43.65%	['fuzz_detokenize', 'fuzz_split_xml', 'fuzz_normalizer', 'fuzz_tokenizer']

This section shows which files and directories are considered in this report. The main reason for showing this is fuzz introspector may include more code in the reasoning than is desired. This section helps identify if too many files/directories are included, e.g. third party code, which may be irrelevant for the threat model. In the event too much is included, fuzz introspector supports a configuration file that can exclude data from the report. See the following link for more information on how to create a config file: link

Files in report

Source file	Reached by	Covered by
...fuzz_tokenizer	['fuzz_tokenizer']	[]
click	[]	[]
xml	[]	[]
itertools	[]	[]
sacremoses.indic	[]	[]
tqdm	[]	[]
sacremoses.cli	['fuzz_detokenize']	[]
copy	[]	[]
atheris	[]	[]
sacremoses.__main__	[]	[]
	[]	[]
sacremoses.normalize	['fuzz_detokenize', 'fuzz_normalizer']	[]
sacremoses.subwords	[]	[]
joblib	[]	[]
sacremoses.sent_tokenize	[]	[]
...fuzz_split_xml	['fuzz_split_xml']	[]
...fuzz_normalizer	['fuzz_normalizer']	[]
sacremoses.chinese	[]	[]
pkgutil	[]	[]
sacremoses.truecase	['fuzz_detokenize', 'fuzz_split_xml']	[]
sacremoses.corpus	['fuzz_tokenizer', 'fuzz_detokenize']	[]
regex	[]	[]
warnings	[]	[]
re	[]	[]
sacremoses.tokenize	['fuzz_tokenizer', 'fuzz_detokenize']	[]
io	[]	[]
sacremoses.util	['fuzz_detokenize']	[]
sacremoses	[]	[]
collections	[]	[]
functools	[]	[]
os	[]	[]
...fuzz_detokenize	['fuzz_detokenize']	[]

Directories in report

Directory