semgrep: the future of static analysis tools

When searching for a pattern that might be present in source code contained in multiple files, what is the best tool to use?

The obvious answer is grep, and grep is great for character-based pattern searches. But patterns that are token based, or include information on language semantics, fall outside grep‘s model of pattern recognition (which does not stop people trying to cobble something together, perhaps with the help of complicated sed scripts).

Those searching source code written in C have the luxury of being able to use Coccinelle, an industrial strength C language aware pattern matching tool. It is widely used by the Linux kernel maintainers and people researching complicated source code patterns.

Over the 15+ years that Coccinelle has been available, there has been a lot of talk about supporting other languages, but nothing ever materialized.

About six months ago, I noticed semgrep and thought it interesting enough to add to my list of tool bookmarks. Then, a few days ago, I read a brief blog post that was interesting enough for me to check out other posts at that site, and this one by Yoann Padioleau really caught my attention. Yoann worked on Coccinelle, and we had an interesting email exchange some 13-years ago, when I was analyzing if-statement usage, and had subsequently worked on various static analysis tools, and was now working on semgrep. Most static analysis tools are created by somebody spending a year or so working on the implementation, making all the usual mistakes, before abandoning it to go off and do other things. High quality tools come from people with experience, who have invested lots of time learning their trade.

The documentation contains lots of examples, and working on the assumption that things would be a lot like using Coccinelle, I jumped straight in.

The pattern I choose to search for, using semgrep, involved counting the number of clauses contained in Python if-statement conditionals, e.g., the condition in: if a==1 and b==2: contains two clauses (i.e., a==1, b==2). My interest in this usage comes from ideas about if-statement nesting depth and clause complexity. The intended use case of semgrep is security researchers checking for vulnerabilities in code, but I’m sure those developing it are happy for source code researchers to use it.

As always, I first tried building the source on the Github repo, (note: the Makefile expects a git clone install, not an unzipped directory), but got fed up with having to incrementally discover and install lots of dependencies (like Coccinelle, the code is written on OCaml {93k+ lines} and Python {13k+ lines}). I joined the unwashed masses and used pip install.

The pattern rules have a yaml structure, specifying the rule name, language(s), message to output when a match is found, and the pattern to search for.

After sorting out various finger problems, writing C rather than Python, and misunderstanding the semgrep output (some of which feels like internal developer output, rather than tool user developer output), I had a set of working patterns.

The following two patterns match if-statements containing a single clause (if.subexpr-1), and two clauses (if.subexpr-2). The option commutative_boolop is set to true to allow the matching process to treat Python’s or/and as commutative, which they are not, but it reduces the number of rules that need to be written to handle all the cases when ordering of these operators is not relevant (rules+test).

rules:
- id: if.subexpr-1
  languages: [python]
  message: if-cond1
  patterns:
   - pattern: |
      if $COND1:  # we found an if statement
         $BODY
   - pattern-not: |
      if $COND2 or $COND3: # must not contain more than one condition
         $BODY
   - pattern-not: |
      if $COND2 and $COND3:
         $BODY
  severity: INFO

- id: if.subexpr-2
  languages: [python]
  options:
   commutative_boolop: true # Reduce combinatorial explosion of rules
  message: if-cond2
  pattern-either:
   - patterns:
      - pattern: |
         if $COND1 or $COND2: # if statement containing two conditions
            $BODY
      - pattern-not: |
         if $COND3 or $COND4 or $COND5: # must not contain more than two conditions
            $BODY
      - pattern-not: |
         if $COND3 or $COND4 and $COND5:
            $BODY
   - patterns:
      - pattern: |
         if $COND1 and $COND2:
            $BODY
      - pattern-not: |
         if $COND3 and $COND4 and $COND5:
            $BODY
      - pattern-not: |
         if $COND3 and $COND4 or $COND5:
            $BODY
  severity: INFO

The rules would be simpler if it were possible for a pattern to not be applied to code that earlier matched another pattern (in my example, one containing more clauses). This functionality is supported by Coccinelle, and I’m sure it will eventually appear in semgrep.

This tool has lots of rough edges, and is still rapidly evolving, I’m using version 0.82, released four days ago. What’s exciting is the support for multiple languages (ten are listed, with experimental support for twelve more, and three in beta). Roughly what happens is that source code is mapped to an abstract syntax tree that is common to all supported languages, which is then pattern matched. Supporting a new language involves writing code to perform the mapping to this common AST.

It’s not too difficult to map different languages to a common AST that contains just tokens, e.g., identifiers and their spelling, literals and their value, and keywords. Many languages use the same operator precedence and associativity as C, plus their own extras, and they tend to share the same kinds of statements; however, declarations can be very diverse, which makes life difficult for supporting a generic AST.

An awful lot of useful things can be done with a tool that is aware of expression/statement syntax and matches at the token level. More refined semantic information (e.g., a variable’s type) can be added in later versions. The extent to which an investment is made to support the various subtleties of a particular language will depend on its economic importance to those involved in supporting semgrep (Return to Corp is a VC backed company).

Outside of a few languages that have established tools doing deep semantic analysis (i.e., C and C++), semgrep has the potential to become the go-to static analysis tool for source code. It will benefit from the network effects of contributions from lots of people each working in one or more languages, taking their semgrep skills and rules from one project to another (with source code language ceasing to be a major issue). Developers using niche languages with poor or no static analysis tool support will add semgrep support for their language because it will be the lowest cost path to accessing an industrial strength tool.

How are the VC backers going to make money from funding the semgrep team? The traditional financial exit for static analysis companies is selling to a much larger company. Why would a large company buy them, when they could just fork the code (other company sales have involved closed-source tools)? Perhaps those involved think they can make money by selling services (assuming semgrep becomes the go-to tool). I have a terrible track record for making business predictions, so I will stick to the technical stuff.

An experiment involving matching regular expressions

Recommendations for/against particular programming constructs have one thing in common: there is no evidence backing up any of the recommendations. Running experiments to measure the impact of particular language features on developer performance is not something that researchers do (there have been a handful of experiments looking at the impact of strong typing on developer performance; the effect measured was tiny).

In February I discovered two groups researching regular expressions. In the first post on duplicate regexs, I promised to say something about the second group. This post discusses an experiment comparing developer comprehension of various regular expressions; the paper is: Exploring Regular Expression Comprehension.

The experiment involved 180 workers on Mechanical Turk (to be accepted, workers had to correctly answer four or five questions about regular expressions). Workers/subjects performed two different tasks, matching and composition.

  • In the matching task workers saw a regex and a list of five strings, and had to specify whether the regex matched (or not) each string (there was also an unsure response).
  • In the composition task workers saw a regular expression, and had to create a string matched by this regex. Each worker saw 10 different regexs, which were randomly drawn from a set of 60 regexs (which had been created to be representative of various regex characteristics). I have not analysed this data yet.

What were the results?

For the matching task: given each of the pairs of regexs below, which one (of each pair) would you say workers were most likely to get correct?

         R1                  R2
1.     tri[a-f]3         tri[abcdef]3
2.     no[w-z]5          no[wxyz]5
3.     no[w-z]5          no(w|x|y|z)5
4.     [ˆ0-9]            [\D]

The percentages correct for (1) were essentially the same, at 94.0 and 93.2 respectively. The percentages for (2) were 93.3 and 87.2, which is odd given that the regex is essentially the same as (1). Is this amount of variability in subject response to be expected? Is the difference caused by letters being much less common in text, so people have had less practice using them (sounds a bit far-fetched, but its all I could think of). The percentages for (3) are virtually identical, at 93.3 and 93.7.

The percentages for (4) were 58 and 73.3, which surprised me. But then I have been using regexs since before \D support was generally available. The MTurk generation have it easy not having to use the ‘hard stuff’ 😉

See Table III in the paper for more results.

This matching data might be analysed using Item Response theory, which can take into account differences in question difficulty and worker/subject ability. The plot below looks complicated, but only because there are so many lines. Each numbered colored line is a different regex, worker ability is on the x-axis (greater ability on the right), and the y-axis is the probability of giving a correct answer (code+data; thanks to Peipei Wang for fixing the bugs in my code):

Probability of giving a correct answer, by subject ability, for 60 regex matching questions

Yes, for question 51 the probability of a correct answer decreases with worker ability. Heads are being scratched about this.

There might be some patterns buried in amongst all those lines, e.g., particular kinds of patterns require a given level of ability to handle, or correct response to some patterns varying over the whole range of abilities. These are research questions, and this is a blog article: answers in the comments :-)

This is the first experiment of its kind, so it is bound to throw up more questions than answers. Are more incorrect responses given for longer regexs, particularly if they cannot be completely held in short-term memory? It is convenient for the author to use a short-hand for a range of characters (e.g., a-f), and I was expecting a difference in performance when all the letters were enumerated (e.g., abcdef); I had theories for either one being less error-prone (I obviously need to get out more).

Patterns of regular expression usage: duplicate regexs

Regular expressions are widely used, but until recently they were rarely studied empirically (i.e., just theory research).

This week I discovered two groups studying regular expression usage in source code. The VTLeeLab has various papers analysing 500K distinct regular expressions, from programs written in eight languages and StackOverflow; Carl Chapman and Peipei Wang have been looking at testing of regular expressions, and also ran an interesting experiment (I will write about this when I have decoded the data).

Regular expressions are interesting, in that their use is likely to be purely driven by an application requirement; the use of an integer literals may be driven by internal housekeeping requirements. The number of times the same regular expression appears in source code provides an insight (I claim) into the number of times different programs are having to solve the same application problem.

The data made available by the VTLeeLab group provides lots of information about each distinct regular expression, but not a count of occurrences in source. My email request for count data received a reply from James Davis within the hour :-)

The plot below (code+data; crates.io has not been included because the number of regexs extracted is much smaller than the other repos) shows the number of unique patterns (y-axis) against the number of identical occurrences of each unique pattern (x-axis), e.g., far left shows number of distinct patterns that occurred once, then the number of distinct patterns that each occur twice, etc; colors show the repositories (language) from which the source was obtained (to extract the regexs), and lines are fitted regression models of the form: NumPatterns = a*MultOccur^b, where: a is driven by the total amount of source processed and the frequency of occurrence of regexs in source, and b is the rate at which duplicates occur.

Number of distinct patterns occurring a given number of times in the source stored in various repositories

So most patterns occur once, and a few patterns occur lots of times (there is a long tail off to the unplotted right).

The following table shows values of b for the various repositories (languages):

StackOverflow   cpan    godoc    maven    npm  packagist   pypi   rubygems
    -1.8        -2.5     -2.5    -2.4    -1.9     -2.6     -2.7     -2.4

The lower (i.e., closer to zero) the value of b, the more often the same regex will appear.

The values are in the region of -2.5, with two exceptions; why might StackOverflow and npm be different? I can imagine lots of duplicates on StackOverflow, but npm (I’m not really familiar with this package ecosystem).

I am pleased to see such good regression fits, and close power law exponents (I would have been happy with an exponential fit, or any other equation; I am interested in a consistent pattern across languages, not the pattern itself).

Some of the code is likely to be cloned, i.e., cut-and-pasted from a function in another package/program. Copy rates as high as 70% have been found. In this case, I don’t think cloned code matters. If a particular regex is needed, what difference does it make whether the code was cloned or written from scratch?

If the same regex appears in source because of the same application requirement, the number of reuses should be correlated across languages (unless different languages are being used to solve different kinds of problems). The plot below shows the correlation between number of occurrences of distinct regexs, for each pair of languages (or rather repos for particular languages; top left is StackOverflow).

Correlation of number of identical pattern occurrences, between pairs of repositories.

Why is there a mix of strong and weakly correlated pairs? Is it because similar application problems tend to be solved using different languages? Or perhaps there are different habits for cut-and-pasted source for developers using different repositories (which will cause some patterns to occur more often, but not others, and have an impact on correlation but not the regression fit).

There are lot of other interesting things that can be done with this data, when connected to the results of the analysis of distinct regexs, but these look like hard work, and I have a book to finish.