What is DiffPlus?

DiffPlus is now used by NAPALM Huawei VRP driver! See PR.

What is DiffPlus?
HOWTO
- Install
- Indented config to dict
- Sanitizing
- Incremental diff
- Diff using a third-party module (optional)
Known limitations
Does DiffPlus reinvent the wheel?

What is DiffPlus?

A lightweight module to help in the comparison of config files. In particular, it computes an incremental diff between two indented config files whilst respecting the scope of the indented blocks (aka contextual diff).

The module leverages the n-ary tree data structure to achieve such a diff. Taking about a hundred lines of code, it only relies on Python builtins and has no extra dependencies.

What for?

The whole point of DiffPlus is to compare config files before running them in production.

Basically, we'd like to merge A config file (the candidate config, partial or full) into another B config file (the running config). The next section states the problem and gives an insight of what we expect from the module.

Problem

Say we have two indented configs:

Config A (to merge into B) Config B

!
no ip domain lookup
!
interface FastEthernet0/0
 no shutdown
!
interface FastEthernet0/0.10
 description LAN
 encapsulation dot1Q 10
 ip address 192.168.1.254 255.255.255.0
!
router bgp 64512
 neighbor 172.16.0.1 remote-as 100
 !
 address-family ipv4
  neighbor 172.16.0.1 activate
  neighbor 172.16.0.1 allowas-in 1
  neighbor 172.16.0.1 prefix-list IN in
  neighbor 172.16.0.1 prefix-list OUT out
  network 192.168.1.0 mask 255.255.255.0
!

!
hostname R1
!
interface FastEthernet0/0
 description LAN
 no ip address
 shutdown
 duplex auto
 speed auto
!
router bgp 64512
 neighbor 172.16.0.1 remote-as 100
 !
 address-family ipv4
  neighbor 172.16.0.1 activate
  neighbor 172.16.0.1 prefix-list IN in
  neighbor 172.16.0.1 prefix-list OUT out
!
ip prefix-list IN seq 5 permit 192.168.2.0/24
ip prefix-list OUT seq 5 permit 192.168.1.0/24
!

The example shows Cisco configs but it applies to any indented config (NOT necessarily network-related).

Some items of A are missing in B. How to find them?

We are only looking for new items to be added in B — it is called an incremental diff
The comparison must respect the indented blocks scope — it is called a contextual diff

A line-by-line diff will not help here since A is typically a partial config to merge into a full one being B.

We humans are able to compute that diff with ease because we visually identify blocks and items (though we make mistakes, sometimes). After some effort, we'd end up with the following result in mind:

# items of A to be added in B (to be computed by diffplus)

+no ip domain lookup
 interface FastEthernet0/0
+ no shutdown
+interface FastEthernet0/0.10
+ description LAN
+ encapsulation dot1Q 10
+ ip address 192.168.1.254 255.255.255.0
 router bgp 64512
  address-family ipv4
+  neighbor 172.16.0.1 allowas-in 1
+  network 192.168.1.0 mask 255.255.255.0

As stated above, this is an incremental diff: there are only additions (hence the module name), no deletions.

As simple as it seems, such a diff is not so trivial for an algorithm.

Proposed solution

Convert each config to an n-ary tree
Do a deep comparison of the n-ary trees

The n-ary tree data structure will help: to represent how nested the items are, to do the match between blocks.

Because we deal with config files, each line is unique per indented block. Therefore, no need for a list of nodes (allowing for duplicates). We can directly use raw nested dicts as n-ary trees, keys being the lines.

For comparison, this gist illustrates another approach based on a list of nodes.

The deep comparison can then be achieved with either a dedicated third-party module or the lightweight IncrementalDiff helper embedded in this module.

HOWTO

The config files used below are configA.txt and configB.txt.

Install

DiffPlus is available on PyPI:

pip install diffplus

Indented config to dict

The to_dict() method converts an indented config to an n-ary tree:

from diffplus import IndentedConfig
from json import dumps

configA = open('configA.txt').read() # or it may be a string
configB = open('configB.txt').read() # or it may be a string

configA = IndentedConfig(configA, comment_char='!', sanitize=True)
configB = IndentedConfig(configB, comment_char='!', sanitize=True)

print(dumps(configA.to_dict(), indent=2))
print(dumps(configB.to_dict(), indent=2))

Output:

Config A (to merge into B)

Config B

{
  "no ip domain lookup": {},
  "interface FastEthernet0/0": {
    "no shutdown": {}
  },
  "interface FastEthernet0/0.10": {
    "description LAN": {},
    "encapsulation dot1Q 10": {},
    "ip address 192.168.1.254 255.255.255.0": {}
  },
  "router bgp 64512": {
    "neighbor 172.16.0.1 remote-as 100": {},
    "address-family ipv4": {
      "neighbor 172.16.0.1 activate": {},
      "neighbor 172.16.0.1 allowas-in 1": {},
      "neighbor 172.16.0.1 prefix-list IN in": {},
      "neighbor 172.16.0.1 prefix-list OUT out": {},
      "network 192.168.1.0 mask 255.255.255.0": {}
    }
  }
}

{
  "hostname R1": {},
  "interface FastEthernet0/0": {
    "description LAN": {},
    "no ip address": {},
    "shutdown": {},
    "duplex auto": {},
    "speed auto": {}
  },
  "router bgp 64512": {
    "neighbor 172.16.0.1 remote-as 100": {},
    "address-family ipv4": {
      "neighbor 172.16.0.1 activate": {},
      "neighbor 172.16.0.1 prefix-list IN in": {},
      "neighbor 172.16.0.1 prefix-list OUT out": {}
    }
  },
  "ip prefix-list IN seq 5 permit 192.168.2.0/24": {},
  "ip prefix-list OUT seq 5 permit 192.168.1.0/24": {}
}

There is no list but only dicts. Each item may have child items. The nesting level is NOT limited.

At this point, you are free to compare the dicts the way you want. However, to that end, you may be interested in existing modules like DeepDiff or the lightweight IncrementalDiff helper embedded in this module.

ℹ Depending on your config format, both indent_char and comment_char can be set at init:

IndentedConfig(config, comment_char='#', indent_char=' ') # default values (e.g., for Huawei)
IndentedConfig(config, comment_char='!')                  # '!' as comment_char (e.g., for Cisco)
IndentedConfig(config, indent_char='\t')                  # tab as indent_char
IndentedConfig(config, sanitize=True)                     # see next section

Sanitizing

Your config may:

have trailing spaces
have blank lines
contain comments
not be correctly indented

The sanitize() method has been made to address this. The last point especially would break the tree conversion. The other ones would make some parts of the diff irrelevant.

from diffplus import IndentedConfig

config = """
   # a global command badly indented
   sysname PE
#
interface GigabitEthernet0/0/1
   # badly indented as well
   description my-super-description
 ip address 1.1.1.1 255.255.255.0
#
"""

config = IndentedConfig(config, comment_char='#')
print(config)
config.sanitize()
print(config)

Print before sanitizing ❌

Print after sanitizing ✔

 
   # a global command badly indented
   sysname PE
#
interface GigabitEthernet0/0/1
   # badly indented as well
   description my-super-description
 ip address 1.1.1.1 255.255.255.0
#

sysname PE
interface GigabitEthernet0/0/1
 description my-super-description
 ip address 1.1.1.1 255.255.255.0

ℹ It is recommended to always sanitize at init:

config = IndentedConfig(config, sanitize=True)

Incremental diff

Diff only

To compute the diff just as introduced in the #problem section:

from diffplus import IndentedConfig, IncrementalDiff

configA = open('configA.txt').read()
configB = open('configB.txt').read()

configA = IndentedConfig(configA, comment_char='!', sanitize=True)
configB = IndentedConfig(configB, comment_char='!', sanitize=True)
diff = IncrementalDiff(configA, configB)

print(diff)

Output:

# items of A to be added in B (computed by IncrementalDiff)

+no ip domain lookup
 interface FastEthernet0/0
+ no shutdown
+interface FastEthernet0/0.10
+ description LAN
+ encapsulation dot1Q 10
+ ip address 192.168.1.254 255.255.255.0
 router bgp 64512
  address-family ipv4
+  neighbor 172.16.0.1 allowas-in 1
+  network 192.168.1.0 mask 255.255.255.0

Merging

Alternatively, we can merge A into B. It is useful to get a preview of the full config before applying it:

from diffplus import IndentedConfig, IncrementalDiff

configA = open('configA.txt').read()
configB = open('configB.txt').read()

configA = IndentedConfig(configA, comment_char='!', sanitize=True)
configB = IndentedConfig(configB, comment_char='!', sanitize=True)
diff = IncrementalDiff(configA, configB, merge=True)

print(diff)

Output:

# items of A merged into B (computed by IncrementalDiff)

 hostname R1
 interface FastEthernet0/0
  description LAN
  no ip address
  shutdown
  duplex auto
  speed auto
+ no shutdown
 router bgp 64512
  neighbor 172.16.0.1 remote-as 100
  address-family ipv4
   neighbor 172.16.0.1 activate
   neighbor 172.16.0.1 prefix-list IN in
   neighbor 172.16.0.1 prefix-list OUT out
+  neighbor 172.16.0.1 allowas-in 1
+  network 192.168.1.0 mask 255.255.255.0
 ip prefix-list IN seq 5 permit 192.168.2.0/24
 ip prefix-list OUT seq 5 permit 192.168.1.0/24
+no ip domain lookup
+interface FastEthernet0/0.10
+ description LAN
+ encapsulation dot1Q 10
+ ip address 192.168.1.254 255.255.255.0

Colored diff

This option is useful to better visualize the changes in the diff:

from diffplus import IndentedConfig, IncrementalDiff

configA = open('configA.txt').read()
configB = open('configB.txt').read()

configA = IndentedConfig(configA, comment_char='!', sanitize=True)
configB = IndentedConfig(configB, comment_char='!', sanitize=True)
diff = IncrementalDiff(configA, configB, merge=True, colored=True)

print(diff)

Output:

Not colored	Colored

Colorization is done through ANSI escape sequences: \033[32m for green color and \033[m for reset color.

Under the hood

The incremental diff is computed recursively as a dict you can access (if needed) through the to_dict() method:

diff = IncrementalDiff(configA, configB, merge=False) # or merge=True
print(dumps(diff.to_dict(), indent=2))

Output:

{
  "+no ip domain lookup": {},
  "interface FastEthernet0/0": {
    "+no shutdown": {}
  },
  "+interface FastEthernet0/0.10": {
    "description LAN": {},
    "encapsulation dot1Q 10": {},
    "ip address 192.168.1.254 255.255.255.0": {}
  },
  "router bgp 64512": {
    "address-family ipv4": {
      "+neighbor 172.16.0.1 allowas-in 1": {},
      "+network 192.168.1.0 mask 255.255.255.0": {}
    }
  }
}

New items are marked with a + so that they can be pretty rendered recursively via the __str__() special method:

print(diff)
print(str(diff)) # equivalent
print(diff.__str__()) # equivalent

Diff using a third-party module

Converting IndentedConfig to dict allows for deep comparison using existing modules like DeepDiff. This way, you are not stuck with the rather simplistic IncrementalDiff helper.

from diffplus import IndentedConfig
from deepdiff import DeepDiff

configA = open('configA.txt').read()
configB = open('configB.txt').read()

configA = IndentedConfig(configA, comment_char='!', sanitize=True)
configB = IndentedConfig(configB, comment_char='!', sanitize=True)
diff = DeepDiff(configB.to_dict(), configA.to_dict())

for item_added in diff['dictionary_item_added']:
    print(item_added)

Output:

root['no ip domain lookup']
root['interface FastEthernet0/0.10']
root['interface FastEthernet0/0']['no shutdown']
root['router bgp 64512']['address-family ipv4']['neighbor 172.16.0.1 allowas-in 1']
root['router bgp 64512']['address-family ipv4']['network 192.168.1.0 mask 255.255.255.0']

But third-party modules are generally heavier (since they offer more features) and may not do exactly what you want. For example, the above output does not include (on purpose) the children of missing items, e.g., interface FastEthernet0/0.10 is missing from B as well as its children yet NOT displayed:

interface FastEthernet0/0.10 # displayed in above output as root['interface FastEthernet0/0.10']
 description LAN                        # not displayed in above output
 encapsulation dot1Q 10                 # not displayed in above output
 ip address 192.168.1.254 255.255.255.0 # not displayed in above output

Known limitations

Only additions, no deletions

Because DiffPlus focuses on merging a config into another one—and not just doing a line-by-line diff—deletions aren't as easy as additions to compute. How to know that an item of A will affect another one in B?

Let's take an example in a network context:

# computed by diffplus (only additions)

 interface FastEthernet0/0
  description Some interface
+ no description
  ip address 10.0.0.1 255.255.255.0
  ip address 10.0.0.2 255.255.255.0 secondary
  ip address 10.0.0.3 255.255.255.0 secondary
+ no ip address 10.0.0.3 255.255.255.0 secondary
+ ip address 10.0.0.4 255.255.255.0 secondary
  speed 10
+ speed 100

Some of the new items will negate or change existing ones. So we'd like a smarter diff:

# NOT computed by diffplus (additions and deletions)

 interface FastEthernet0/0
- description Some interface
  ip address 10.0.0.1 255.255.255.0
  ip address 10.0.0.2 255.255.255.0 secondary
- ip address 10.0.0.3 255.255.255.0 secondary
+ ip address 10.0.0.4 255.255.255.0 secondary
- speed 10
+ speed 100

We humans are able to compute that diff because we visually identify items and are familiar with the config logic. From an algorithmic point of view, however, it is challenging. Not only it depends on the config grammar and syntax (i.e., what are considered keywords and values) but also on the semantic (e.g., adding an item won't necessarily replace a similar one as it is the case for secondary addresses).

The closest string match approach using an helper like difflib.get_close_matches() is an interesting lead, yet not 100% accurate and it would have over-complexified the module.

Code-like config

By essence, DiffPlus is not suited for code diff. Each line is assumed to be unique per indented block. Therefore, the tree conversion won't work well with algorithms as they have statements which repeat in the code.

Let's take an example with pseudocode:

# algo.txt

if some_expression then
 first_if_content
else
 first_else_content
fi
if anoter_expression then
 second_if_content
else
 second_else_content
fi

Tree conversion:

from diffplus import IndentedConfig
from json import dumps

algo = open('algo.txt').read()
algo = IndentedConfig(algo)

print(dumps(algo.to_dict(), indent=4))

Output:

{
    "if some_expression then": {
        "first_if_content": {}
    },
    "else": {
        "second_else_content": {}
    },
    "fi": {},
    "if anoter_expression then": {
        "second_if_content": {}
    }
}

Inconsistencies:

There is only one fi instead of twos.
Likewise, the first else has been overwritten by the second one at the same indentation level.

This is in accordance with how the module works: it has been designed for config diff, not code diff.

One character for indentation

For now, DiffPlus does not support multiple characters as indentation symbol:

IndentedConfig(config, indent_char='  ')   # 2 spaces (NOT supported)
IndentedConfig(config, indent_char='    ') # 4 spaces (NOT supported)
IndentedConfig(config, indent_char='\t')   # tab char (supported)

The first two lines will raise an error:

ValueError: "indent_char" must be a char, not an str

The same limitation applies for comment_char.

A future version may support it if the module gains interest in the community.

Does DiffPlus reinvent the wheel?

I didn't find a suitable module providing such an incremental and contextual diff simply based on an indented config (NOT necessarily Cisco-based or network-based for the sake of genericity).

difflib (Python builtin) does a line-by-line diff
DeepDiff compares dicts but does not convert an indented text to dict
ConfigTree only supports YAML and JSON formats
anytree drops the support of the indented text format
diffios only supports Cisco format and it is not clear what it does exactly (variable parsing or diff?)
shconfparser (probably the closest one) only supports Cisco format and does not offer diff features
conf_diff does a line-by-line diff

Update: netutils, which is partially used in the NAPALM project, provides a similar diff mechanism. Yet network-oriented, the proposed implementation is interesting.

Name		Name	Last commit message	Last commit date
Latest commit History 33 Commits
.github/workflows		.github/workflows
diffplus		diffplus
tests		tests
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
pyproject.toml		pyproject.toml

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

What is DiffPlus?

What for?

Problem

Proposed solution

HOWTO

Install

Indented config to dict

Sanitizing

Incremental diff

Diff only

Merging

Colored diff

Under the hood

Diff using a third-party module

Known limitations

Only additions, no deletions

Code-like config

One character for indentation

Does DiffPlus reinvent the wheel?

About

Releases 5

Languages

License

angely-dev/diffplus

Folders and files

Latest commit

History

Repository files navigation

What is DiffPlus?

What for?

Problem

Proposed solution

HOWTO

Install

Indented config to dict

Sanitizing

Incremental diff

Diff only

Merging

Colored diff

Under the hood

Diff using a third-party module

Known limitations

Only additions, no deletions

Code-like config

One character for indentation

Does DiffPlus reinvent the wheel?

About

Topics

Resources

License

Stars

Watchers

Forks

Releases 5

Languages